Patent Application
20230125295
The following disclosure relates generally to techniques for automatically determining the acquisition location of an image on a building floor plan based on an analysis of visual data of the image relative to analysis of the floor plan and for subsequently using the determined acquisition location information in one or more manners, such as to use the image's acquisition location to improve navigation of the building.
In various fields and circumstances, such as architectural analysis, property inspection, real estate acquisition and development, general contracting, improvement cost estimation, automated navigation, etc., it may be desirable to know the interior of a house, office, or other building without having to physically travel to and enter the building. However, it can be difficult to effectively capture, represent and use such building interior information, including to display visual information captured within building interiors to users at remote locations (e.g., to enable a user to fully understand the layout and other details of the interior, including to control the display in a user-selected manner). In addition, while a floor plan of a building may provide some information about layout and other details of a building interior, such use of floor plans has some drawbacks, including that floor plans can be difficult to construct and maintain, to accurately scale and populate with information about room interiors, to visualize and otherwise use, etc.
The present disclosure describes techniques for using computing devices to perform automated operations related to determining the acquisition location of an image based at least in part on an analysis of visual data of the image and comparison to corresponding analyzed floor plan information, and for subsequently using the determined image acquisition location information in one or more further automated manners. In at least some embodiments, images to be analyzed include one or more panorama images or other images (e.g., rectilinear perspective images) acquired at one or more acquisition locations in an interior of a multi-room building (e.g., a house, office, etc.), and the determined image acquisition location information includes at least a location on a floor plan of the building and in some situations further includes an orientation or other direction information for at least a part of the image(s)—in at least some such embodiments, the automated image acquisition location determination is further performed without having or using any acquired depth data from any depth sensors or other distance-measuring devices about distances from an image's acquisition location to walls or other objects in the surrounding building. The determined image acquisition location information may be further used in various manners in various embodiments, such as in conjunction with a corresponding building floor plan and/or other generated mapping-related information (e.g., a three-dimensional model of the building's interior), including for controlling navigation of mobile devices (e.g., autonomous vehicles), for display or other presentation on one or more client devices in corresponding GUIs (graphical user interfaces), etc. Additional details are included below regarding the automated acquisition and use of determined image acquisition location information, and some or all of the techniques described herein may, in at least some embodiments, be performed via automated operations of an Image Floor Plan Location Mapping Manager (“IFPLMM”) system, as discussed further below.
In at least some embodiments and situations, some or all of the images acquired for a building are each a panorama image that is acquired at an acquisition location in or around the building, such as to generate a panorama image at each of multiple such acquisition locations from one or more of a video at that acquisition location (e.g., a 360° video taken from a smartphone or other mobile device held by a user turning at that acquisition location), or multiple images acquired in multiple directions from the acquisition location (e.g., from a smartphone or other mobile device held by a user turning at that acquisition location), or a simultaneous capture of all the image information (e.g., using one or more fisheye lenses), etc. It will be appreciated that such a panorama image may in some situations be presented using an equirectangular projection (with vertical lines and other vertical information in an environment being shown as straight lines in the projection, and with horizontal lines and other horizontal information in the environment being shown in the projection in a curved manner if they are above or below a horizontal centerline of the image and with an amount of curvature increasing as a distance from the horizontal centerline increases) and provide up to 360° coverage around horizontal and/or vertical axes, such that a user viewing a starting panorama image may move the viewing direction within the starting panorama image to different orientations to cause different images (or “views”) to be rendered within the starting panorama image (including to present the image being rendered as a perspective image using a planar coordinate system). Furthermore, acquisition metadata regarding the capture of such panorama images may be obtained and used in various manners in some embodiments, such as data acquired from IMU (inertial measurement unit) sensors or other sensors of a mobile device as it is carried by a user or otherwise moved between acquisition locations, while in other embodiments such acquisition metadata is not acquired or used (e.g., so as to determine an acquired image's acquisition location on a building floor plan based solely on the visual data of the image). Additional details are included below related to the acquisition and usage of panorama images or other images for a building.
As noted above, automated operations of an IFPLMM system may include determining the acquisition location of an image that is captured in a defined area of a building (e.g., in a room of a house or other building) based at least in part on an analysis of the visual information included in the image's contents. In at least some embodiments, such automated determination of an image's acquisition location may include some or all of the following: using a trained neural network (e.g., a deep convolutional neural network) to analyze the image and determine features of the visual data of the image contents (e.g., based on colors and/or estimated depth for each pixel; based on information for groups of associated pixels, such as colors and/or textures and/or estimated depths; etc., or more generally to determine one or more latent space visual data features that are generated by the trained neural network); creating a circular descriptor for the image that has a quantity of degrees corresponding to a field of view of the image (e.g., a circular descriptor that represents 360 degrees for a panorama image with 360° of horizontal visual coverage, a circular descriptor that represents 180° for a panorama image with 180° of horizontal visual coverage, a circular descriptor that represents 72 degrees for a non-panorama perspective image with 72° of horizontal visual coverage, etc.) and with each degree of the image circular descriptor encoding information about determined feature(s) from the image for a corresponding degree in the image's visual data (e.g., relative to a direction within the image that is designated as 0° or otherwise as a starting angle or direction); and comparing the image circular descriptor to one or more corresponding building location circular descriptors that each represents information from the building's floor plan for a corresponding location associated with the building, with one or more best matching building location circular descriptors then identified and used to determine an acquisition location of the image in a room or other area associated with the building.
Consider, for the purposes of an illustrative example, a panorama image captured in a room of a building, with the panorama image including 360° of horizontal coverage around a vertical axis (e.g., a full circle showing all of the walls of the room from the acquisition location of the panorama image, unless a portion of a wall is occluded by intervening objects in the room and/or by another intervening wall, such as for room shapes that are not purely rectangular), and with the x and y axes of the image's visual contents being aligned with corresponding horizontal and vertical information in the room (e.g., the border between two walls, the border between a wall and the floor, the bottoms and/or tops of windows and doors, etc.), such that the image is not skewed or otherwise misaligned with respect to the room. For the purposes of this example, the image capture may be performed sequentially at multiple directions from an acquisition location using changing camera orientations, beginning with a camera orientation in a northern direction that corresponds to a relative starting horizontal direction of 0° for this panorama image, and continues in a circle, with a relative 90° horizontal direction for this panorama image then corresponding to the eastern direction, a relative 180° horizontal direction for this panorama image corresponding to the southern direction, a relative 270° horizontal direction for this panorama image corresponding to the western direction, and a relative 360° ending horizontal direction for this panorama image being back to the northern direction. In at least some embodiments, the information about the locations of identified features of the visual data of the panorama image are encoded in a manner specific to such angular degrees of direction from the acquisition location (e.g., relative to the starting direction of the panorama image), producing an image circular descriptor for the image that encodes information about 360° of visual coverage in the image—thus, the image circular descriptor for such an image may encode information about what features are identified in each of 360 horizontal degrees (e.g., to combine feature information for a given horizontal degree direction for a range of vertical degree directions in that horizontal degree direction). Such information about the locations of identified features may be encoded and stored in various manners in various embodiments, including in some embodiments in an array or vector having one or more values for each angular degree of direction, such as to identify one or more features present in a given angular direction. In addition, in other embodiments and situations, the feature information for an image may be identified and represented in manners other than based on angular differences from a starting direction of the image, resulting in other types of image descriptors that are used in similar manners. Additional details are included below regarding the construction and use of such image circular descriptors, including with respect to the examples of
As noted above, the automated determination of the acquisition location of an image in a room of a building (or other defined building area) by an embodiment of the IFPLMM system may include matching the angular information encoded in the image's generated image circular descriptor to corresponding angular information in building location circular descriptors generated from a floor plan for the building, such as to determine a particular image acquisition location (and optionally orientation) in the room or other area. In order to generate the building location circular descriptors from a building's floor plan, the automated operations of the IFPLMM system may further include starting with an existing floor plan of the building, such as a two-dimensional rasterized floor plan that uses a top-down view to show at least the wall structures of the building interior and optionally has additional associated information, such as semantic information about structural wall elements (e.g., locations of doors, windows, and other inter-room wall openings) and/or structural information for additional areas associated with the building (e.g., one or more additional exterior structures, such as a separate garage or car port, accessory dwelling unit, shed, etc.; one or more other external areas, such as a yard, garden, porch, deck, patio, balcony, sidewalk or other path, etc.), and generating a point cloud corresponding to at least the structural information of the floor plan (e.g., a two-dimensional, or “2D”, point cloud; a three-dimensional, or “3D”, point cloud if additional height information is available; etc.)—each point in the point cloud may further be assigned one or more types of associated information, such as a 2D location on the floor plan (e.g., a relative location from a designated spot, such as to treat the lowest and left-most point as 0,0 for X-Y axes), a normal direction (e.g., relative to a planar surface formed by that point and adjacent points), associated semantic information (if any) associated with that floor plan position, etc. In at least some embodiments, such automated determination of one or more building location circular descriptors for the floor plan may include some or all of the following: using a trained neural network (e.g., a deep convolutional neural network, such as similar to but separately trained from the neural network used to analyze images' visual data) to analyze the floor plan point cloud data and any additional associated information to determine corresponding features (e.g., one or more latent space data features generated by the trained neural network); and, for each room or other area associated with the building, creating one or more building location circular descriptors for that room or other area (e.g., multiple building location circular descriptors for multiple locations within that room or other area, such as in a grid or other chosen formation) that each has a selected quantity of degrees (e.g., 360°) and with each degree of a building location circular descriptor encoding information about any determined feature(s) from the floor plan and associated information for any floor plan point cloud points present at a corresponding degree direction in the floor plan from that building location (e.g., with a direction designated as 0° or otherwise as a starting angle or direction, such as corresponding to north or to an upward direction on a 2D floor plan). Such building location circular descriptors may be predetermined, for example, before any corresponding image circular descriptors are generated or used, or may instead in some situations be dynamically created at a time of use for comparison to an image circular descriptor for an image taken in an area associated with the building. In some embodiments, the feature information for a particular degree in a building location circular descriptor for a building location may encode information about an incident angle from the building location to any point(s) of the point cloud in the direction of that particular degree from the building location (e.g., using an enumerated group of incident angle ranges, such as relative to the point(s)′ normal direction) and an estimated distance between the building location and the location(s) of the point(s) (e.g., using an enumerated group of distance ranges). In addition, in at least some embodiments, additional information may be associated with the building location circular descriptor(s) generated for a building room or other area and used for comparison of an image circular descriptor for a new image, such as to also use and encode visual data in the building location circular descriptor(s) from one or more other images that have previously been localized within that room or other area (localized by determining at least the acquisition location of such an image within that room or other area and by optionally further determining an orientation of the image within that room or other area, so as to together specify the ‘pose’ of the image within the room or other area, and such as with the orientation optionally identified by associating a particular image angle or other direction with a corresponding floor plan angle or other direction for the room or other area so as to align different rotational coordinate systems of the image and the floor plan, such as with the orientation optionally identified by determining a geographical direction, such as north, to which directions within the image and floor plan correspond, etc.).
Once a plurality of building location circular descriptors are generated or otherwise obtained for a plurality of room locations in a room or other area of a building, they may be compared or otherwise matched to an image circular descriptor for an image taken in the room or other area in order to determine at least one of the building location circular descriptors that is a best match, with the acquisition location of the image then being identified based at least in part on the room location of the best match building location circular descriptor(s). For example, the image's determined acquisition location may in some embodiments and situations be selected to be that room location for a best match building location circular descriptor, or instead in other embodiments and situations the image's determined acquisition location may be determined to be within a small distance from that room location (e.g., in a direction and/or amount based on differences between the image circular descriptor and that best match building location circular descriptor), such as by using a trained neural network (e.g., separate from the one or more neural networks used to determine the features for the image and building floor plan point cloud) to refine the image acquisition location to a position between the building locations having defined building location circular descriptors. The matching process for an image circular descriptor and a building location circular descriptor may include determining a distance and/or or amount of similarity/dissimilarity between the two circular descriptors in one or more manners, including in a rotation-independent manner, such as by determining the probability that two circular descriptors are matching (with the highest matching probability corresponding to the smallest dissimilarity and/or distance), by measuring the differences between the vectors or other encoded formats for the circular descriptors being compared, etc.—as one non-exclusive example, a circular earth mover's distance metric may be used to compare the vectors for two such circular descriptors in a rotation-independent manner (e.g., regardless of whether the two circular descriptors use the same direction in the room as their respective relative starting points), while in other embodiments differences in rotation between two descriptors may be handled in other manners. In addition, the matching process may in some embodiments include comparing the image circular descriptor to each possible building location circular descriptor (e.g., all building location circular descriptors generated for one or more candidate buildings in which the image may have been captured), while in other embodiments only a subset of the building location circular descriptors for a particular building may be considered (e.g., by performing a nearest neighbor gradient ascent or descent search using a defined similarity or dissimilarity metric). Additional details are included below regarding the construction and use of such building location circular descriptors, including for comparison to one or more image circular descriptors, such as with respect to the examples of
The described techniques provide various benefits in various embodiments, including to allow floor plans of multi-room buildings and other structures to be automatically augmented with information about acquisition locations at which images are acquired in or around the buildings or other structures, including without having or using information from depth sensors or other distance-measuring devices about distances from images' acquisition locations to walls or other objects in a surrounding building or other structure—in at least some such embodiments, the determination of an image's acquisition location in an area associated with one or more buildings is further performed without having or using a predicted room layout from the image and/or without having any other images that are previously registered with determined acquisition locations on a building's floor plan. Furthermore, such automated techniques allow such image acquisition location information to be determined more quickly than previously existing techniques, and in at least some embodiments with greater accuracy, including by using information acquired from the actual building environment (rather than from plans on how the building should theoretically be constructed), as well as enabling the capture of changes to structural elements that occur after a building is initially constructed if a corresponding building floor plan reflects that actual building environment and/or such changes. Such described techniques further provide benefits in allowing improved automated navigation of a building by mobile devices (e.g., semi-autonomous or fully-autonomous vehicles), based at least in part on the determined acquisition locations of images, including to significantly reduce computing power and time used to attempt to otherwise learn a building's layout. In addition, in some embodiments the described techniques may be used to provide an improved GUI in which a user may more accurately and quickly obtain information about a building's interior (e.g., for use in navigating that interior), including in response to search requests, as part of providing personalized information to the user, as part of providing value estimates and/or other information about a building to a user, etc. Various other benefits are also provided by the described techniques, some of which are further described elsewhere herein.
As noted above, automated operations of an IFPLMM system may include determining the acquisition location of an image that is taken in a defined area (e.g., in a room of a house or other building) based at least in part on an analysis of the visual information included in the image's contents. In at least some embodiments, such an IFPLMM system may operate in conjunction with one or more separate ICA (Image Capture and Analysis) systems and/or with one or more separate MIGM (Mapping Information and Generation Manager) systems, such as to obtain and use floor plans and other associated information for buildings from the MIGM system and/or to obtain images to be localized for a building from the ICA system, while in other embodiments such an IFPLMM system may incorporate some or all functionality of such ICA and/or MIGM systems as part of the IFPLMM system. In yet other embodiments, the IFPLMM system may operate without using some or all functionality of the ICA and/or MIGM systems, such as if the IFPLMM system obtains information about building floor plans and/or other associated information from other sources (e.g., from manual creation by one or more users, from provision of such building floor plans and/or associated information by one or more external systems or other sources, etc.), and/or if the IFPLMM system obtains information about images to be localized from other sources (e.g., from end users, such as in a crowdsourced manner). In addition, building floor plans that are used in the manner described herein may be in various formats (whether as originally obtained and/or after an initial automated analysis by the IFPLMM system), including in at least some embodiments to be in a vectorized form with specified information about the locations of structural elements such as one or more of the following: walls, windows, doorways and other inter-room openings, corners, etc. (e.g., after initially receiving a non-vectorized image form of the building floor plan that is analyzed to produce the vectorized form).
With respect to functionality of such an ICA system, it may perform automated operations in at least some embodiments to acquire one or more images (e.g., panorama images) at one or more acquisition locations associated with a building (e.g., in the interior of multiple rooms of the building, at one or more exterior locations, etc.), and optionally further acquire metadata related to the image acquisition process and/or to movement of a capture device between multiple acquisition locations. For example, in at least some such embodiments, such techniques may include using one or more mobile devices (e.g., a camera having one or more fisheye lenses and mounted on a rotatable tripod or otherwise having an automated rotation mechanism; a camera having one or more fisheye lenses sufficient to capture 360° horizontally without rotation; a smart phone held and moved by a user, such as to rotate the user's body and held smart phone in a 360° circle around a vertical axis; a camera held by or mounted on a user or the user's clothing; a camera mounted on an aerial and/or ground-based drone or robotic device; etc.) to capture visual data from one or more acquisition locations but without acquiring information from any depth sensors or other distance-measuring devices about distances between the acquisition location(s) and objects in an environment around the acquisition location(s), such as from a sequence of multiple acquisition locations within multiple rooms of a house (or other building). Additional details are included elsewhere herein regarding operations of device(s) implementing an ICA system, such as to perform such automated operations, and in some cases to further interact with one or more ICA system operator user(s) in one or more manners to provide further functionality.
With respect to functionality of such an MIGM system, it may perform automated operations in at least some embodiments to analyze multiple 360° panorama images (and optionally other images) that have been acquired for a building interior (and optionally an exterior of the building), and determine shapes or rooms or other areas and locations of passages connecting rooms or other areas for some or all of those panorama images, as well as to determine wall elements and other elements of some or all rooms or other areas of the building in at least some embodiments and situations. The types of connecting passages between two or more rooms or other areas may include one or more of doorway openings and other inter-room non-doorway wall openings, windows, stairways, non-room hallways, etc., and the automated analysis of the images may identify such elements based at least in part on identifying the outlines of the passages, identifying different content within the passages than outside them (e.g., different colors or shading), etc. The automated operations may further include using the determined information to generate a floor plan for the building and to optionally generate other mapping information for the building (e.g., a 3D model of the building interior and/or exterior), such as by using the inter-room passage information and other information to determine relative positions of the associated shapes of rooms or other areas to each other, and to optionally add distance scaling information and/or various other types of information to the generated floor plan. In addition, the MIGM system may in at least some embodiments perform further automated operations to determine and associate additional information with a building floor plan and/or specific rooms, areas or locations within the floor plan, such as to analyze images and/or other environmental information (e.g., audio) captured within the building interior to determine particular attributes (e.g., a color and/or material type and/or other characteristics of particular elements, such as a floor, wall, ceiling, countertop, furniture, fixtures, appliances, etc.; the presence and/or absence of particular elements, such as an island in the kitchen; etc.), or to otherwise determine relevant attributes (e.g., directions that building elements face, such as windows; views from particular windows or other locations; etc.). Additional details are included below regarding operations of computing device(s) implementing an MIGM system, such as to perform such automated operations and in some cases to further interact with one or more MIGM system operator user(s) in one or more manners to provide further functionality.
The described techniques provide various benefits in various embodiments, including to improve the control of autonomous operations of excavator construction vehicles and/or other construction vehicles (e.g., fully autonomous operations), such as based at least in part on training one or more machine learning behavioral models for one or more such construction vehicles (e.g., one or more construction vehicle types) and on using the trained machine learning behavior model(s) to control corresponding autonomous operations of one or more corresponding construction vehicles. Furthermore, such automated techniques allow such training and usage operations to be performed more quickly and with greater accuracy than previously existing techniques, including to significantly reduce computing power and time used. In addition, in some embodiments the described techniques may be used to provide an improved GUI in which a user may more accurately and quickly obtain information about operations of excavator construction vehicles and/or other construction vehicles, including in response to search requests or other instructions, as part of providing personalized information to the user, etc. Various other benefits are also provided by the described techniques, some of which are further described elsewhere.
For illustrative purposes, some embodiments are described below in which specific types of information are acquired, used and/or presented in specific ways for specific types of structures and by using specific types of devices—however, it will be understood that the described techniques may be used in other manners in other embodiments, and that the invention is thus not limited to the exemplary details provided. As one non-exclusive example, while specific types of circular descriptors are generated for images and for room locations and are compared or otherwise matched in specific manners in some embodiments, it will be appreciated that other types of information to describe image contents and room locations may be similarly generated and used in other embodiments, including for buildings (or other structures or layouts) separate from houses, and that determined image acquisition location information may be used in other manners in other embodiments. In addition, the term “building” refers herein to any partially or fully enclosed structure, typically but not necessarily encompassing one or more rooms that visually or otherwise divide the interior space of the structure—non-limiting examples of such buildings include houses, apartment buildings or individual apartments therein, condominiums, office buildings, commercial buildings or other wholesale and retail structures (e.g., shopping malls, department stores, warehouses, etc.), etc. The term “acquire” or “capture” as used herein with reference to a building interior, acquisition location, or other location (unless context clearly indicates otherwise) may refer to any recording, storage, or logging of media, sensor data, and/or other information related to spatial characteristics and/or visual characteristics and/or otherwise perceivable characteristics of the building interior or subsets thereof, such as by a recording device or by another device that receives information from the recording device. As used herein, the term “panorama image” may refer to a visual representation that is based on, includes or is separable into multiple discrete component images originating from a substantially similar physical location in different directions and that depicts a larger field of view than any of the discrete component images depict individually, including (but not limited to) images with a sufficiently wide-angle view from a physical location to include angles beyond that perceivable from a person's gaze in a single direction. The term “sequence” of acquisition locations, as used herein, refers generally to two or more acquisition locations that are each visited at least once in a corresponding order, whether or not other non-acquisition locations are visited between them, and whether or not the visits to the acquisition locations occur during a single continuous period of time or at multiple different times, or by a single user and/or device or by multiple different users and/or devices. In addition, various details are provided in the drawings and text for exemplary purposes, but are not intended to limit the scope of the invention. For example, sizes and relative positions of elements in the drawings are not necessarily drawn to scale, with some details omitted and/or provided with greater prominence (e.g., via size and positioning) to enhance legibility and/or clarity. Furthermore, identical reference numbers may be used in the drawings to identify the same or similar elements or acts.
In particular,
In this example, each building location circular descriptor uses the northward direction to correspond to 0°, continuing in a clockwise manner for 360°.
The building location circular descriptor for a given room location may be generated in a variety of manners in various embodiments, including by using geometric techniques to determine the angular amount from a given room location and starting direction to a given location of a point cloud point (e.g., on a wall). For purposes of illustration, a variety of point cloud points 233a-233q are illustrated for room 260a, with corresponding feature information being illustrated in example building location circular descriptor 278c5—while the 360 degrees are represented in a linear fashion in the example building location circular descriptor 278c5 shown in
While not illustrated in the examples of
With respect to finding a best match building location circular descriptor for image circular descriptor 279d from multiple possible building locations in the room, some or all of the building location circular descriptors for the room locations in the grid may be compared to the image circular descriptor 279d to determine a degree of match in various manners in various embodiments. For example, in some embodiments a circular earthmover's distance metric may be used to compare two such descriptors in a rotation independent manner, such that the two descriptors may have relative 0° starting directions that point in different directions. Other measures of distance or similarity/dissimilarity may be used in other embodiments, such as by measuring the distance for each angular degree and aggregating that information across all of the angular degrees.
In addition, to facilitate a comparison of two such circular descriptors in situations in which the distance or similarity/dissimilarity metric is not rotation-independent, additional automated operations may be performed in some embodiments to ensure that information encoded in a given relative angular direction in one circular descriptor is being compared to a relative angular direction in the other circular descriptor that points in the same actual real-world direction. For example, in some embodiments, a brute force method could be used that compares each angular direction in one circular descriptor to a particular angular direction (e.g., the starting direction) in the other circular descriptor, thus ensuring that one of the comparisons uses the same directions. Alternatively, in other embodiments automated operations may be performed to synchronize the two circular descriptors to be compared, such as by identifying which relative angular directions in one circular descriptor correspond to which relative angular directions in the other circular descriptor (e.g., to identify, for the relative 0° starting angular direction for one circular descriptor, what the corresponding angular direction is in the other circular descriptor). With respect to the example of
Various other types of information are also illustrated on the 2D floor plan 255f in this example. For example, such other types of information may include one or more of the following: room labels added to some or all rooms (e.g., “living room” for the living room); room dimensions added for some or all rooms; visual indications of fixtures or appliances or other built-in features added for some or all rooms; visual indications added for some or all rooms of positions of additional types of associated and linked information (e.g., of other panorama images and/or perspective images that an end user may select for further display, of audio annotations and/or sound recordings that an end user may select for further presentation, etc.); visual indications added for some or all rooms of doors and windows; etc. In addition, in this example a user-selectable control 228 is added to indicate a current floor that is displayed for the floor plan, and to allow the end user to select a different floor to be displayed—in some embodiments, a change in floors or other levels may also be made directly from the floor plan, such as via selection of a corresponding connecting passage in the illustrated floor plan (e.g., the stairs to floor 2). It will be appreciated that a variety of other types of information may be added in some embodiments, that some of the illustrated types of information may not be provided in some embodiments, and that visual indications of and user selections of linked and associated information may be displayed and selected in other manners in other embodiments.
In addition, while at least some rooms of the house are represented with associated nodes in the adjacency graph, in at least some embodiments, some spaces within the house may not be treated as rooms for the purpose of the adjacency graph (i.e., may not have separate nodes in the adjacency graph), such as for closets, small areas such as a pantry or a cupboard, connecting areas such as stairs and/or hallways, etc.—in this example embodiment, the stairs have a corresponding node 245h and a walk-in closet may optionally have a node 245l, while the pantry does not have a node, although none or all or any combination of those spaces may have nodes in other embodiments. In addition, in this example embodiment, areas outside of the building that are adjacent to building entries/exits also have nodes to represent them, such as node 245j corresponding to the front yard (which is accessible from the building by the entry door), and node 245i corresponding to the deck (which is accessible by the back door), as well as nodes 245l for an additional exterior structure and 245m for a larger backyard and/or patio—in other embodiments, such external areas may not be represented as nodes (and instead may be represented in some embodiments as attributes associated with adjacent exterior doors or other openings and/or with their rooms). Similarly, in this example embodiment, information about areas that are visible from windows or from other building locations may also be represented by nodes, such as optional node 245k corresponding to the view accessible from the western window in the living room, although in other embodiments such views may not be represented as nodes (and instead may be represented in some embodiments as attributes associated with the corresponding window or other building location and/or with their rooms). It will be noted that, while some edges are shown in
In addition, further automated operations may be performed in at least some embodiments as part of an automated determination of the acquisition location of an image captured in a room or other building area, such as if corresponding information about a building's room or other area is available. For example, in at least some embodiments, a geometric localization technique may be used to test associations of wall elements visible in an image to wall elements present in a room, whether to confirm a degree of match for a building location circular descriptor that has already been determined to be a best match for an image circular descriptor and/or as part of the identification of such a best match building location circular descriptor. The geometric localization technique may include, for example, determining one or more likely room shapes of a room and/or positions of elements within the room using 2-point solvers and/or 3-point solvers (or alternatively receiving such information as a starting point), and then positioning the wall elements on the likely room shape(s)—in other embodiments, the wall element locations may be determined in other manners, such as via use of depth sensing equipment or other room mapping sensors in the room, via a machine learning approach for analysis of images to identify room shapes and wall element locations, via input specified by one or more human operators, etc. Furthermore, in some embodiments, given a room location and information about a room shape and the locations of wall elements, a new synthetic image that is a projection/visualization of a view of the room from that room location may be generated with the wall elements shown in their locations, and the visual information of that synthetic image may be directly compared to the actual image from the room to determine a degree of similarity/dissimilarity or other degree of match between the two images, with that inter-image comparison used to determine if that room location is a match for the acquisition location of the actual image. In a similar manner, in some embodiments, some or all of the building location circular descriptors for room locations in a room may be generated as image circular descriptors of images (e.g., 360° panorama images) taken at those room locations, and those room/image circular descriptors may then be compared to an image circular descriptor of a new image taken in the room (e.g., an image with less than 360° of horizontal coverage) to determine a best match building location circular descriptor in a manner similar to that discussed above.
The automated determination of the acquisition location of an image taken in a room may further include additional operations in some embodiments. For example, in at least some embodiments, machine learning techniques may be used to learn the best encoding to allow matching of an image to a room location, such as from among multiple defined candidate encodings, or instead by considering a variety of possible features and identifying a subset of those image features that provide best matches to corresponding room locations. Additional details are included below regarding various automated operations that may be performed by the IFPLMM system in at least some embodiments.
Additional details related to embodiments of a system providing at least some such functionality of an IFPLMM system or related system for determining images' acquisition locations on floor plans and optionally presenting corresponding information are included in co-pending U.S. Non-Provisional patent application Ser. No. 17/013,323, filed Sep. 4, 2020 and entitled “Automated Analysis Of Image Contents To Determine The Acquisition Location Of The Image” (which includes disclosure of an example ILMM system that is generally directed to automated operations for determining acquisition locations of images acquired in and around a building); and in co-pending U.S. Non-Provisional patent application Ser. No. 17/201,996, filed Mar. 15, 2021 and entitled “Automated Determination Of Image Acquisition Locations In Building Interiors Using Determined Room Shapes” (which includes disclosure of an example ILDM system that is generally directed to automated operations for determining acquisition locations of images acquired in and around a building); each of which is incorporated herein by reference in its entirety.
As one non-exclusive example embodiment, determining a target image's acquisition location within a room (also referred to as ‘localization’ for the purpose of this example embodiment) may include estimating 2D camera pose p*, associated with target image I, with respect to a reference structural layout (or ‘map’) M. Camera pose p=[t, θ]ϵSE(2) is modeled within a 2D plane, having a rotation θ[0, 2π) in yaw axis and a translation t=[x, y], with the pose parameters tϵR2 and θϵ[0,2 π) defining, respectively, the camera's planar displacement vector and yaw axis rotation. The target image may be either a panorama image (e.g., in equirectangular format) or a perspective image with a known Field of View (FoV). Map M can have various formats, and encodes the structural layout information (also referred to as ‘occupancy’ for the purpose of this example embodiment) within a 2D plane.
The determination of the target image's acquisition location in this example embodiment includes using a Monte Carlo Localization (MCL) framework, which defines a measurement model P(I|p; M) that expresses the likelihood of image I observed at camera pose p on a map M, with the non-random parameter M being excluded below for simplicity. The posterior distribution of p after observing I is the solution of interest, and the MCL framework estimates the posterior distribution P (p|I) based on the Bayes' formula as follows:
where P(I) is a normalization constant that can be ignored, and P(p) is the prior camera pose distribution on the map, which is assumed to be uniformly distributed within the map area. Finally, the full posterior can be approximated by drawing particles from P (p) whose likelihoods will be estimated using the measurement model as defined in Equation (5) below.
Under the metric learning framework, instead of using a flattened descriptor used in metric learning, a circular feature is used to encode spatial visibility, leading to the metric learning being geometrically interpretable. A circular feature is defined as an ordered set of feature vectors as follows:
F={f
α|α=1 . . . V−1} (2)
where V is the number of feature segments. Each feature segment fαϵRD encodes a local directional FoV of 2π/V rads in the range (2πα/V) to ((2π*(α+1))/V) on the 2D plane. This ordered set F is referred to as a circular feature in this example embodiment since the first and last feature segments correspond to adjacent FoVs. The ordered feature segments correspond to the 360° 2D spatial information, and V is the number of segments. With this design, the omni-directional 2D spatial information is implicitly encoded in the order of feature segments.
A similarity measurement between two circular features Fi={fiα|α=0 . . . V−1} and Fj={fjα|α=0 . . . V−1} is defined as follows:
where cos(⋅,⋅) computes the vector cosine similarity, and the function output is normalized to [0, 1]. A rotation operator (F, θ) is defined that rotates the underlying spatial information of a circular feature F with a given angle θ by
where the 1D feature space is linearly interpolated when indexing yields non-integer values. Finally, the measurement model is defined as
P(I|p)=P(I|t,θ)=A(
1,
(
t,θ)) (5)
where A is the PDF normalization constant, and FI and Ft are circular features encoded from the target image and rendered at location t on the map respectively.
To systematically reduce the rotation dimension from the MCL sampling step (since MCL uses a large number of samples to approximate the camera pose posterior in SE(2)), and for a sample location t with a canonical orientation, the optimal relative rotation of its circular feature Ft with respect to an image circular feature FI can be found by
Substitute to Equation 5 provides a simplified measurement model obviating 6 and that is only conditioned on translation t as
P(I|t)=A′·(
I,
(
t,θtopen)) (7)
For solving Equation 6, Ft is rotated with uniformly sampled θt in [0, 2π), and the best is kept. This discretized search initializes rotation to a rough value, which will later be refined as discussed below. The rotation matching process is highly efficient since it reuses the same circular feature and does not render new hypotheses.
Circular features are rendered from 2D floor maps with a given camera pose in the following manner. Given a general 2D map representation M (e.g., floor plan, occupancy grid, etc.) that encodes the area occupancy information, points are uniformly sampled on the occupancy boundaries (e.g., walls) to extract a 2D point cloud set M={mi|i=0 . . . N−1}. Each point mi=[ti, ni, si] encodes its location ti, normal vector ni and an optional semantic information si. The normal vector is normalized pointing inside the room. When semantic information (e.g., labels and/or positions for doors, windows, etc.) is available, it is encoded as binary masks appended to the point representation.
To circumvent inefficient two-stage render-and-encode processes, a latent space rendering is used that directly renders circular feature vectors by aggregating features from visible map points. However, visibility for static environments is locally constant at most sampling locations, providing a limited spatial context, so detailed rendering dynamics (such as length and incident-angle of the viewing rays between features and sampling location) are analyzed to mitigate potential homogenization of the representations, and an adaptive rendering mechanism is defined. Hence, a rendering codebook is used that corresponds to an over-specified latent feature space in order to endow map points with view-dependent adaptive features encoding rendering dynamics (e.g., using the features of map points during rendering as determined by the codebook and rendering dynamics, with features from multiple codebooks mixed with addition). The 2D point cloud map is processed (e.g., with a PointNet) to assign each map point mi two sets of latent features Gi={giβ|β=0 . . . G−1} and Hi={hiY|γ=0 . . . H−1}, which is denoted as distance and incident-angle codebook respectively. Features in the codebook have the same dimension as circular feature segments gβ, hγϵRD. At rendering time, the map point features are chosen from the codebooks based on their distance and incident-angle with respect to the rendering location. Assuming a rendering location {circumflex over (t)} and a map point mi=[ti, ni, si], let di=ti−{circumflex over (t)}, its rendering dynamics are computed as
d
i
=∥d
i∥ (8)
ψi=a tan 2(∥di×ni∥,di·ni) (9)
where di and ψi are distance and incident-angle respectively. The clockwise incident-angle [0, 2π) distinguishes the four quadrants. With m/s associated codebooks Gi and Hi, its feature fi is then determined by
where dmax is a predefined maximum distance for the distance codebook. Similar to Equation 4, for non-integer indexing, linear interpolation between its two closest codes in the codebook may be used. Finally, if mi passes the visibility test to location {circumflex over (t)}, fi is projected to circular feature F{circumflex over (t)}={fiα|α=0 . . . V−1} by
where ωi=a tan 2(di) is the angle of viewing ray. Finally, the projected map point features are averaged into each segment.
Circular features are extracted from panorama and/or perspective images in the following manner. For panorama images in equirectangular projection, each image column in the image corresponds to a fixed amount of horizontal FoV. Such capture configuration facilitates a bijective mapping between groups of adjacent input image columns and the segments in the rendered circular representation. The query panorama is fed into an encoder (e.g., a ResNet50 encoder) to obtain a feature map, which is subsequently squeezed by averaged pooling in the vertical dimension to comply with the feature dimensions of the circular segments, and averaged-pooled again in the horizontal direction to V elements, in accordance with the preconfigured number of feature segments in each circular feature.
Perspective images are the most common images that taken from cameras subject to the pinhole camera model, which has additional degrees of freedoms in camera rotation compared to panorama images. If an image is assumed to have known FoV and zero pitch/roll angle related to the ground plane, each image column in the perspective image will correspond to a non-fixed but known amount of horizontal FoV—note that for indoor target images, the pitch/roll angle can usually be rectified by estimating the room layout. The target image may be processed to a feature map (e.g., using a ResNet50 encoder), and average pooling may be used to squeeze the vertical dimension, with a perspective-to-equirectangular transform on the feature map to get the final circular feature. Since perspective images have significantly less than 360° FoV, their circular features will have segments without assigned values, which will be masked out in the computation. Equation 3 can also be re-normalized to have range [0, 1].
The accuracy of the Monte-Carlo localization framework can depend on the sampling density, but it can be inefficient to achieve a high accuracy using sampling. A lightweight continuous refinement branch is used to address the discretized pose sampling nature of the MCL processing, to improve upon a current estimation. With current best estimation t* and θ*, the refinement branch takes two circular features FI and (Ft*, θ*) as input. The refinement network uses two 1D convolution layers with circular padding followed by a fully-connected layer to predict two offsets δt, δθ for translation and rotation respectively. The updated map circular feature Ft*+δt, θ*+δθ is then rendered, and its similarity is computed to FI using Equation 3. If the similarity score improved upon the original camera pose, it is accepted and iteration continues (e.g., with converging often occurring within 3 iterations, and with the first refinement accepted in at least some situations to unquantize the estimation), and otherwise the refinement is treated as being converged.
Triplet loss may be used as supervision for learning the metric space. To form a triplet, the image circular feature FI is used as the anchor, the map circular feature at ground truth camera pose F+=(Ftgt, θgt) is used as a positive feature and a negative map circular feature F−=
(Ftmd, θmd) is used at an outlier camera pose. Then the triplet loss is defined by
triplet=2·max((
I,
+)−
(
I,
−)+0.5,0) (12)
The similarity function S and triplet loss Ltriplet uses aggregated element-wise comparisons, and thus effectively disregards any intra-feature context. An additional context loss can be used to provide feature segments a global scope of their circular features for learning context information (e.g., properties of the room/map). Circular feature context F is defined as the mean of its normalized feature segments
which is applied to training triplets in a manner similar to Equation 12.
context=max(cos(I,
+)−cos(
I,
−)+1.0,0) (14)
With the context loss, the circular features achieve better coarse level expressiveness, improving recall for target images with limited FoV. This loss also acts as a regularizer by mitigating feature segments having large variance, leading to a smoother posterior estimation.
For training the refinement branch, circular features within a 0.5 meter radius and a 30 degree angle from the ground truth camera pose are sampled, and the refinement branch is supervised using a regression loss as
refine_t=∥(tgt−t*)−δt∥
refine_r=min(|(θgt−θ*)−δθ|,2π−|(θgt−θ*)−δθ|) (15)
For triplet and context loss, 100 negative samples are sampled and broadcast the single ground truth sample for each training iteration. For refinement loss, 20 hard negatives are sampled near the ground truth camera pose with a disturbance sampled from uniform distribution bounded in 30 degree and 0.5 meter radius. The mean of all losses is combined with equal weights, and hyper-parameters are set as G=H=32, V=16, D=128 and dmax=10 m consistently through benchmarking. The map is sampled into a 2D point cloud with a 10 cm interval at occupancy boundaries. Circular features are rendered for a 0.1 m×0.1 m uniform grid within the map range. For estimating relative rotations as described in Equation 6, 16 uniformly sampled angle are evaluated and the best is kept. Finally, the posterior distribution can be estimated using Equations 1 and 7. To extract final estimations from the posterior grid map, a 3×3 non-maximum suppression is applied to extract the maximums. For maximums that have a larger score than a threshold (e.g., 0.8), they are sent into the refinement branch to get the final estimations with their likelihoods as uncertainty estimation. Sorting by their likelihoods, a top-k estimation is available. While this example embodiment discussed above may in some situations use a binary occlusion status with respect to a viewing ray from a map location, other embodiments and/or situations may further support partial occlusion, such as by modeling map objects (e.g., furniture). In addition, while this example embodiment discussed above may in some situations not use information about whether doors are open or not, other embodiments and/or situations may utilize such information. Furthermore, while this example embodiment discussed above may in some situations use 2D floor maps, other embodiments and/or situations may be extended to 3D space and/or model complex rendering-time dynamics such as non-Lambertian reflection (e.g., mirrors).
Various details have been provided with respect to
In addition,
One or more users (not shown) of one or more client computing devices 175 may further interact over one or more computer networks 170 with the IFPLMM system 140 and optionally the ICA system 160 and/or MIGM system 160, such as to obtain and use determined acquisition location information for images (e.g., to obtain and view and optionally interact with one or more such images and/or a generated floor plan on which the one or more images have been located, such as to optionally perform one or more of the following: to change between a floor plan view and a view of a particular image at an acquisition location within or near the floor plan; to change the horizontal and/or vertical viewing direction from which a corresponding view of a panorama image is displayed, such as to determine a portion of a panorama image to which a current user viewing direction is directed, etc.). In addition, while not illustrated in
In the depicted computing environment of
In the example of
In operation, a user associated with the mobile device arrives at a first acquisition location 210A within a first room of the building interior (in this example, an entryway from an external door 190-1 to the living room), and captures a view of a portion of the building interior that is visible from that acquisition location 210A (e.g., some or all of the first room, and optionally small portions of one or more other adjacent or nearby rooms, such as through doors, halls, stairs or other connecting passages from the first room) as the mobile device is rotated around a vertical axis at the first acquisition location (e.g., with the user turning his or her body in a circle while holding the mobile device stationary relative to the user's body). The actions of the user and/or the mobile device may be controlled or facilitated via use of one or more programs executing on the mobile device, such as ICA application system 154, optional browser 162, control system 147, etc., and the view capture may be performed by recording a video and/or taking a succession of one or more images, including to capture visual information depicting a number of objects or other elements (e.g., structural details) that may be visible in images (e.g., video frames) captured from the acquisition location. In the example of
With respect to such a captured panorama image, the user may also optionally in some embodiments provide a textual or auditory identifier to be associated with the panorama image and/or its acquisition location, such as “entry” for acquisition location 210A or “living room” for acquisition location 210B, while in other embodiments the ICA system may automatically generate such identifiers (e.g., by automatically analyzing video and/or other recorded information for a building to perform a corresponding automated determination, such as by using machine learning) or the identifiers may not be used. After the first acquisition location 210A has been adequately captured (e.g., by a full rotation of the mobile device), the user may optionally proceed to a next acquisition location (such as acquisition location 210B), optionally recording movement data during movement between the acquisition locations, such as video and/or other data from the hardware components (e.g., from one or more IMUs, from the camera, etc.). At the next acquisition location, the user may similarly use the mobile device to capture one or more images from that acquisition location. This process may repeat from some or all rooms of the building and optionally external to the building, as illustrated for acquisition locations 210C-210L, including in this example to capture one or more panorama images on an external deck or patio or balcony 186, to capture one or more panorama images on a larger external yard or patio 187, to capture one or more panorama images near or in an external additional structure 188 (e.g., a garage, shed, accessory dwelling unit, greenhouse, etc.). The acquired video and/or other images for each acquisition location are further analyzed to generate a panorama image for each of acquisition locations 210A-210L, including in some embodiments to match objects and other elements in different images. In addition to generating such panorama images, further analysis may be performed in order to ‘link’ at least some of the panoramas together (with some corresponding lines 215 between them being shown for the sake of illustration), such as to determine relative positional information between pairs of acquisition locations that are visible to each other, to store corresponding inter-panorama links (e.g., links 215-AB, 215-BC and 215-AC between acquisition locations A and B, B and C, and A and C, respectively), and in some embodiments and situations to further link at least some acquisition locations that are not visible to each other (e.g., a link 215-BE, not shown, between acquisition locations 210B and 210E).
Additional details related to embodiments of generating and using linking information between panorama images, including using travel path information and/or elements or other features visible in multiple images, are included in U.S. Non-Provisional patent application Ser. No. 16/693,286, filed Nov. 23, 2019 and entitled “Connecting And Using Building Data Acquired From Mobile Devices” (which includes disclosure of an example BICA system that is generally directed to obtaining and using linking information to inter-connect multiple panorama images captured within one or more buildings or other structures), in U.S. Non-Provisional patent application Ser. No. 17/080,604, filed Oct. 26, 2020 and entitled “Generating Floor Maps For Buildings From Automated Analysis Of Visual Data Of The Buildings' Interiors”; and in U.S. Provisional Patent Application No. 63/035,619, filed Jun. 5, 2020 and entitled “Automated Generation On Mobile Devices Of Panorama Images For Buildings Locations And Subsequent Use”; each of which is incorporated herein by reference in its entirety. In addition, additional details related to embodiments of a system providing at least some such functionality of an MIGM system or related system for generating floor plans and associated information and/or presenting floor plans and associated information are included in U.S. Non-Provisional patent application Ser. No. 16/190,162, filed Nov. 14, 2018 and entitled “Automated Mapping Information Generation From Inter-Connected Images” (which includes disclosure of an example Floor Map Generation Manager, or FMGM, system that is generally directed to automated operations for generating and displaying a floor plan or other floor plan of a building using images acquired in and around the building); in U.S. Non-Provisional patent application Ser. No. 16/681,787, filed Nov. 12, 2019 and entitled “Presenting Integrated Building Information Using Three-Dimensional Building Models” (which includes disclosure of an example FMGM system that is generally directed to automated operations for displaying a floor plan or other floor plan of a building and associated information); in U.S. Non-Provisional patent application Ser. No. 16/841,581, filed Apr. 6, 2020 and entitled “Providing Simulated Lighting Information For Three-Dimensional Building Models” (which includes disclosure of an example FMGM system that is generally directed to automated operations for displaying a floor plan or other floor plan of a building and associated information); in U.S. Non-Provisional patent application Ser. No. 17/080,604, filed Oct. 26, 2020 and entitled “Generating floor plans For Buildings From Automated Analysis Of Visual Data Of The Buildings' Interiors” (which includes disclosure of an example VTFM system that is generally directed to automated operations for generating a floor plan or other floor plan of a building using visual data acquired in and around the building); and in U.S. Non-Provisional patent application Ser. No. 16/807,135, filed Mar. 2, 2020 and entitled “Automated Tools For Generating Mapping Information For Buildings” (which includes disclosure of an example MIGM system that is generally directed to automated operations for generating a floor plan or other floor plan of a building using images acquired in and around the building); and in U.S. Non-Provisional patent application Ser. No. 17/069,800, filed Oct. 13, 2020 and entitled “Automated Tools For Generating Building Mapping Information” (which includes disclosure of an example MIGM system that is generally directed to automated operations for generating mapping information for a building using images acquired in and around the building); each of which is incorporated herein by reference in its entirety. Moreover, further details related to embodiments of a system providing at least some such functionality of a system for using acquired images and/or generated floor plans are included in U.S. Non-Provisional patent application Ser. No. 17/185,793, filed Feb. 25, 2021 and entitled “Automated Usability Assessment Of Buildings Using Visual Data Of Captured In-Room Images” (which includes disclosure of an example Building Usability Assessment Manager, or BUAM, system that is generally directed to automated operations for analyzing visual data from images captured in rooms of a building to assess room layout and other usability information for the building's rooms and optionally for the overall building, and to subsequently using the assessed usability information in one or more further automated manners); each of which is incorporated herein by reference in its entirety.
Various details are provided with respect to
The server computing system(s) 380 and executing IFPLMM system 389 may communicate with other computing systems and devices via one or more networks 399 (e.g., the Internet, one or more cellular telephone networks, etc.), such as user client computing devices 390 (e.g., used to view floor plans, associated images and/or other related information), ICA and MIGM server computing system(s) 300, one or more mobile image acquisition devices 360, optionally other navigable devices 395 that receive and use floor plans and determined image acquisition locations and optionally other generated information for navigation purposes (e.g., for use by semi-autonomous or fully autonomous vehicles or other devices), and optionally other computing systems that are not shown (e.g., used to store and provide additional information related to buildings; used to capture building interior data; used to store and provide information to client computing devices, such as additional supplemental information associated with images and their encompassing buildings or other surrounding environment; etc.).
In the illustrated embodiment, an embodiment of the IFPLMM system 389 executes in memory 387 in order to perform at least some of the described techniques, such as by using the processor(s) 381 to execute software instructions of the system 389 in a manner that configures the processor(s) 381 and computing system 380 to perform automated operations that implement those described techniques. The illustrated embodiment of the IFPLMM system may include one or more components, not shown, to each perform portions of the functionality of the IFPLMM system, and the memory may further optionally execute one or more other programs 388—as one specific example, copies of the ICA and/or MIGM systems may execute as one of the other programs 388 in at least some embodiments, such as instead of or in addition to the ICA system 340 and MIGM system 345 on the server computing system(s) 300. The IFPLMM system 389 may further, during its operation, store and/or retrieve various types of data on storage 385 (e.g., in one or more databases or other data structures), such as various types of floor plan information and other building mapping information 391 (e.g., generated and saved 2D floor plans and semantic information about positions of wall elements and other elements on those floor plans, generated and saved 2.5D and/or 3D models, building and room dimensions for use with associated floor plans, additional images and/or annotation information, etc.), information 393 about images whose acquisition locations are to be determined and associated information 392 about such determined acquisition locations, information 394 about generated building location circular descriptors and image circular descriptors, and optionally various other types of information. The ICA system 340 and/or MIGM system 345, if present, may similarly store and/or retrieve various types of data on storage 320 (e.g., in one or more databases or other data structures) during their operation and provide some or all such information to the IFPLMM system 389 for its use (whether in a push and/or pull manner), such as various types of floor plan information and other building mapping information 326 (e.g., similar to or the same as information 391), various types of user information 322, acquired 360° panorama image information 324 (e.g., for analysis to generate floor plans; to provide to users of client computing devices 390 for display; etc., optionally including information about inter-panorama image links that reflect relative positional information for the panorama images), and/or various types of optional additional information 329 (e.g., various analytical information related to presentation or other use of one or more building interiors or other environments captured by an ICA system).
Some or all of the user client computing devices 390 (e.g., mobile devices), mobile image acquisition devices 360, other navigable devices 395 and other computing systems may similarly include some or all of the same types of components illustrated for server computing systems 300 and 380. As one non-limiting example, the mobile image acquisition devices 360 are each shown to include one or more hardware CPU(s) 361, I/O components 362, storage 365, imaging system 364, IMU hardware sensors 369, and memory 367, with one or both of a browser 368 and one or more client applications 369 (e.g., an application specific to the IFPLMM system and/or ICA system) executing within memory 367, such as to participate in communication with the IFPLMM system 389, ICA system 340 and/or other computing systems. While particular components are not illustrated for the other navigable devices 395 or client computing systems 390, it will be appreciated that they may include similar and/or additional components.
It will also be appreciated that computing systems 300 and 380 and the other systems and devices included within
It will also be appreciated that, while various items are illustrated as being stored in memory or on storage while being used, these items or portions of them may be transferred between memory and other storage devices for purposes of memory management and data integrity. Alternatively, in other embodiments some or all of the software components and/or systems may execute in memory on another device and communicate with the illustrated computing systems via inter-computer communication. Thus, in some embodiments, some or all of the described techniques may be performed by hardware means that include one or more processors and/or memory and/or storage when configured by one or more software programs (e.g., by the IFPLMM system 389 executing on server computing systems 380) and/or data structures, such as by execution of software instructions of the one or more software programs and/or by storage of such software instructions and/or data structures, and such as to perform algorithms as described in the flow charts and other disclosure herein. Furthermore, in some embodiments, some or all of the systems and/or components may be implemented or provided in other manners, such as by consisting of one or more means that are implemented partially or fully in firmware and/or hardware (e.g., rather than as a means implemented in whole or in part by software instructions that configure a particular CPU or other processor), including, but not limited to, one or more application-specific integrated circuits (ASICs), standard integrated circuits, controllers (e.g., by executing appropriate instructions, and including microcontrollers and/or embedded controllers), field-programmable gate arrays (FPGAs), complex programmable logic devices (CPLDs), etc. Some or all of the components, systems and data structures may also be stored (e.g., as software instructions or structured data) on a non-transitory computer-readable storage mediums, such as a hard disk or flash drive or other non-volatile storage device, volatile or non-volatile memory (e.g., RAM or flash RAM), a network storage device, or a portable media article (e.g., a DVD disk, a CD disk, an optical disk, a flash memory device, etc.) to be read by an appropriate drive or via an appropriate connection. The systems, components and data structures may also in some embodiments be transmitted via generated data signals (e.g., as part of a carrier wave or other analog or digital propagated signal) on a variety of computer-readable transmission mediums, including wireless-based and wired/cable-based mediums, and may take a variety of forms (e.g., as part of a single or multiplexed analog signal, or as multiple discrete digital packets or frames). Such computer program products may also take other forms in other embodiments. Accordingly, embodiments of the present disclosure may be practiced with other computer system configurations.
The illustrated embodiment of the routine begins at block 405, where information or instructions are received. The routine continues to block 410 to determine whether the instructions or other information received in block 405 indicate to determine building location circular descriptors for an indicated building, and if not continues to block 440. Otherwise, the routine continues to perform blocks 415-430 to determine the building location circular descriptors, including to obtain a rasterized floor plan for the building in block 415 (e.g., to retrieve from storage, to receive in block 405, etc.) that optionally has associated semantic information for structural wall elements (e.g., doors and other inter-room or inter-area openings, windows, inter-room or other inter-area borders, etc.). In block 420, the routine then generates a point cloud (e.g., a 2D point cloud) corresponding to the building's structural elements shown on the floor plan, and determines associated information for each point (e.g., a 2D XY location, normal direction, semantic data, etc.), and generates latent space features for each point using a trained neural network. In block 425, the routine then generates and stores a building location circular descriptor for each of multiple building locations that identifies features of points in directions from that location (e.g., for each of 360 horizontal degrees). In block 430 the routine then optionally generates and stores a graph with nodes that represent rooms or other building areas, and with inter-node edge corresponding to inter-area connectivity or other inter-area adjacency, and associates each node with the generated building location circular descriptors that correspond to building locations in that room or other building area.
After block 430, or if it is instead determined in block 410 that the instructions or other information received in block 405 are not to determine building location circular descriptors for an indicated building, the routine continues to block 440 to determine whether the instructions or other information received in block 405 indicate to determine the acquisition location of an indicated image (e.g., for an indicated building or within any known buildings), and if so the routine continues to perform blocks 445-485 to do so, and otherwise continues to block 490.
In block 445, the routine obtains information about the image whose acquisition location is to be determined, such as by receiving that image in block 405 or by otherwise retrieving a stored copy of the image. In block 450, the routine then proceeds to generate an image circular descriptor for the image that includes information about features of the image's visual data in each of a plurality of angular directions (e.g., at each of 360 horizontal degrees of angular direction, if the image is a 360° panorama image, such as relative to an angular direction determined to be a starting direction for the image).
In block 460, the routine then compares the image circular descriptor to some or all of building location circular descriptors previously generated for one or more buildings (e.g., for an indicated building, such as with respect to all rooms and/or all non-room areas, for one or more rooms and/or non-room areas to which the image may correspond, etc.) to determine a best matching building location circular descriptor, such as a building location circular descriptor having a smallest dissimilarity distance to the image circular descriptor. The routine further identifies the room location to use as the determined acquisition location for the image based on the room location associated with the best match building location circular descriptor, such as to use that associated room location as the determined acquisition location, or to instead optionally do further refinement of the determined acquisition location. In some embodiments and situations, the routine may further determine orientation and/or direction information from that determined acquisition location that corresponds to one or more parts of the image (e.g., to a starting direction for the image and/or to an ending direction for the image). In block 485, the routine then optionally adds information to a graph node corresponding to that determined acquisition location about the image, and uses visual data of the image and its determined acquisition location to update some or all building location circular descriptors in the room or other area associated with that graph node.
After block 485, the routine continues to block 488 to store the information that was determined and generated in blocks 415 to 485, and to optionally display the determined image acquisition location information for the image in its enclosing room or other area on the floor map (or a floor map excerpt), although in other embodiments the determined information may be used in other manners (e.g., for automated navigation of one or more devices).
If it is instead determined in block 440 that the information or instructions received in block 405 are not to determine the acquisition location of an image, the routine continues instead to block 490 to perform one or more other indicated operations as appropriate. Such other operations may include, for example, receiving and responding to requests for previously determined image acquisition location information and/or for associated images (e.g., requests for such information for display on one or more client devices, requests for such information to provide it to one or more other devices for use in automated navigation, etc.), obtaining and storing information about buildings for use in later operations (e.g., information about floor plans and associated wall element positions for rooms in the floor plan, etc.), performing geometric localization techniques to test associations of wall elements visible in an image to wall elements present in a room (whether to confirm a degree of match for a building location circular descriptor that has already been determined to be a best match for an image circular descriptor and/or as part of the identification of such a best match building location circular descriptor), using machine learning techniques to learn the best encoding to allow matching of an image to a room location, etc.
After blocks 488 or 490, the routine continues to block 495 to determine whether to continue, such as until an explicit indication to terminate is received, or instead only if an explicit indication to continue is received. If it is determined to continue, the routine returns to block 405 to wait for and receive additional instructions or information, and otherwise continues to block 499 and ends.
The illustrated embodiment of the routine begins at block 505, where instructions or information are received. At block 510, the routine determines whether the received instructions or information indicate to display or otherwise present information representing one or more building areas (e.g., a building interior), and if not continues to block 590. Otherwise, the routine proceeds to block 512 to retrieve a floor plan and/or other generated mapping information (e.g., a 3D computer model) for the building and optionally indications of associated linked information for the building interior and/or a surrounding location, and selects an initial view of the retrieved information (e.g., a view of the floor plan, of at least some of the 3D computer model, etc.). In block 515, the routine then displays or otherwise presents the current view of the retrieved information, and waits in block 517 for a user selection. After a user selection in block 517, if it is determined in block 520 that the user selection corresponds to the current location (e.g., to change the current view), the routine continues to block 522 to update the current view in accordance with the user selection, and then returns to block 515 to update the displayed or otherwise presented information accordingly. The user selection and corresponding updating of the current view may include, for example, displaying or otherwise presenting a piece of associated linked information that the user selects (e.g., a particular image associated with a displayed visual indication of a determined acquisition location), changing how the current view is displayed (e.g., zooming in or out; rotating information if appropriate; selecting a new portion of the floor plan and/or 3D computer model to be displayed or otherwise presented, such as with some or all of the new portion not being previously visible, or instead with the new portion being a subset of the previously visible information; etc.).
If it is instead determined in block 510 that the instructions or other information received in block 505 are not to present information representing a building interior, the routine continues instead to block 590 to perform any other indicated operations as appropriate, such as any housekeeping tasks, to configure parameters to be used in various operations of the system (e.g., based at least in part on information specified by a user of the system, such as a user of a mobile device who captures one or more building interiors, an operator user of the IFPLMM system, etc.), to obtain and store other information about users of the system, to respond to requests for generated and stored information, etc.
Following block 590, or if it is determined in block 520 that the user selection does not correspond to the current location, the routine proceeds to block 595 to determine whether to continue, such as until an explicit indication to terminate is received, or instead only if an explicit indication to continue is received. If it is determined to continue (e.g., if the user made a selection in block 517 related to a new location to present), the routine returns to block 505 to await additional instructions or information (or to continue on to block 512 if the user made a selection in block 517 related to a new location to present), and if not proceeds to step 599 and ends.
Non-exclusive example embodiments described herein are further described in the following clauses.
A01. A computer-implemented method comprising:
obtaining, by one or more computing devices, and for a house with multiple rooms, a rasterized two-dimensional floor plan of the house that has associated semantic information about locations of doors and windows and inter-wall borders of the multiple rooms;
generating, by the one or more computing devices, building location description information for the house, including:
generating, by the one or more computing devices, an image circular descriptor for a panorama image that is taken in one of the multiple rooms and has 360 horizontal degrees of visual information, including determining second latent space features associated with visual data of the panorama image by supplying the panorama image to a second trained neural network, and wherein the image circular descriptor encodes information identifying specified directions within the visual data to the second latent space features;
comparing, by the one or more computing devices, the image circular descriptor to the building location circular descriptors to determine one of the building location circular descriptors whose encoded information best matches the encoded information of the image circular descriptor;
associating, by the one or more computing devices and based on the comparing, the panorama image with a determined position on the two-dimensional floor plan, wherein the determined position includes the building location in the one room associated with the determined one building location circular descriptor and further includes orientation information to correlate the determined angular directions for that building location to the identified specified directions for the panorama image; and
using, by the one or more computing devices, the determined position of the panorama image on the two-dimensional floor plan of the house for navigation of at least the one room of the house.
A02. The computer-implemented method of clause A01 wherein the generating of the building location circular descriptors further includes obtaining a first enumerated group of ranges of incident angles, obtaining a second enumerated group of ranges of distances, and performing the encoding for each of the building location circular descriptors of the information about some of the first latent space features by, for each of the at least some points for the building location of that building location circular descriptor, encoding information in that building location circular descriptor for one of the 360 horizontal degrees from that building location to that point that includes one of the ranges of incident angles from the first enumerated group and one of the ranges of distances from the second enumerated group.
A03. The computer-implemented method of any one of clauses A01-A02 further comprising using, by the one or more computing devices, the two-dimensional floor plan to further control navigation activities by an autonomous vehicle, including providing the two-dimensional floor plan for use by the autonomous vehicle in moving between the multiple rooms of the house.
A04. The computer-implemented method of any one of clauses A01-A03 wherein the using of the determined position further includes displaying, by the one or more computing devices, the two-dimensional floor plan showing the multiple rooms and including one or more visual indications on the displayed two-dimensional floor plan of the determined position and the orientation information for the panorama image in the one room.
A05. A computer-implemented method comprising:
obtaining, by a computing device and for a building, building location description information including a plurality of building location circular descriptors for a plurality of building locations in the building, wherein each building location circular descriptor is associated with one of the building locations and has first angular information about first latent space features identified for structural elements of the building at specified angular directions from the associated building location, wherein the first latent space features are identified by a first trained neural network using a two-dimensional floor plan of the building;
generating, by the computing device, an image circular descriptor for a panorama image that is captured in a room of the building and that includes visual information about at least some walls of the room, wherein the image circular descriptor has second angular information about second latent space features identified from the visual information of the panorama image at specified directions by a second trained neural network;
comparing, by the computing device, the image circular descriptor to the building location circular descriptors to determine one of the building location circular descriptors that is in the room and has first angular information best matching the second angular information of the image circular descriptor;
associating, by the computing device and based on the comparing, the panorama image with a determined position and orientation in the room, the determined position based on the building location with which the determined one building location circular descriptor is associated, and the determined orientation identifying at least one direction from that building location corresponding to a specified part of the visible information in the panorama image; and
presenting, by the computing device, information that includes the two-dimensional floor plan of the building and shows the room with a visual indication identifying at least the determined position for the panorama image, to cause use of the presented information for navigating the building.
A06. The computer-implemented method of any one of clauses A01-A05 wherein the presenting of the floor plan further includes visually indicating the determined orientation, and wherein the method further comprises presenting, by the computing device and in response to a user selection of the visual indication on the presented floor plan, at least a portion of the panorama image corresponding to the determined orientation.
A07. The computer-implemented method of any one of clauses A01-A06 wherein the visual information of the panorama image includes 360 horizontal degrees of visual coverage from an acquisition location of the panorama image,
wherein the image circular descriptor includes, for each of the 360 horizontal degrees of visual coverage from the acquisition location, information about at least some of the second latent space features associated with any structural elements of the room that are visible in a direction from the acquisition location corresponding to the horizontal degree of visual coverage, and
wherein each of the building location circular descriptors includes, for each of 360 horizontal degrees from the building location associated with the building location circular descriptor, information about at least some of the first latent space features associated with any structural elements of a surrounding room that are visible in a direction from the that building location corresponding to the horizontal degree of visual coverage.
A08. The computer-implemented method of clause A07 wherein the structural elements of the building include at least one door, at least one window, and at least one inter-wall border, and wherein the obtaining of the building location description information includes generating the building location circular descriptors, including generating from the two-dimensional floor plan a two-dimensional point cloud having a plurality of points, including associating information with each of the points that includes two-dimensional location information for the point and normal direction information for the point and semantic information about any structural elements associated with the point, and including analyzing the points and the associated information to generate the first latent space features, wherein each of the points is associated with at least one of the first latent space features.
A09. The computer-implemented method of any one of clauses A07-A08 further comprising determining the one building location circular descriptor having angular information best matching the information included in the image circular descriptor by performing the generating and the comparing without using any depth information acquired from any depth sensor about a depth from the acquisition location to any surrounding elements of the room.
A10. The computer-implemented method of any one of clauses A07-A09 further comprising selecting the plurality of building locations in the building by specifying a grid of building locations covering floors of at least some rooms of multiple rooms of the building.
A11. The computer-implemented method of clause A10 wherein the comparing of the image circular descriptor to the building location circular descriptors includes performing a nearest-neighbor search of the building locations of the grid, including identifying the determined one building location circular descriptor by repeatedly moving from at least one current building location in the grid to at least one neighbor building location in the grid if the at least one neighbor building location has a smaller dissimilarity with the image circular descriptor than does the at least one current building location.
A12. The computer-implemented method of any one of clauses A07-A11 wherein the comparing of the image circular descriptor to the building location circular descriptors further includes:
analyzing the visual information to identify, for a characteristic of a specified type, at least one of the 360 horizontal degrees of visual coverage from the acquisition location for which the characteristic is present;
for each of at least some of the building location circular descriptors, comparing the image circular descriptor to the building location circular descriptor by:
selecting one of the at least some building location circular descriptors as the determined one building location circular descriptor based on the selected one building location circular descriptor having an identified synchronized location for which the information at the other horizontal degrees of coverage in the building location circular descriptor best matches the information at the other horizontal degrees of coverage in the image circular descriptor, and using the identified synchronized location to determine the orientation in the room for the panorama image.
A13. The computer-implemented method of clause A12 wherein the characteristic of the specified type is one of a visible wall being orthogonal to a line along an identified horizontal degree of visual coverage, or a specified type of wall element being visible at the identified horizontal degree of visual coverage.
A14. The computer-implemented method of any one of clauses A01-A13 wherein the comparing of the image circular descriptor to the building location circular descriptors includes, for each of at least some of the building location circular descriptors, determining a probability that the image circular descriptor and the building location circular descriptor are a match by differing less than a specified threshold, and selecting one of the at least some building location circular descriptors that has a highest probability of matching the image angular detector as the determined one building location circular descriptor.
A15. The computer-implemented method of any one of clauses A01-A14 wherein the comparing of the image circular descriptor to the building location circular descriptors includes, for each of at least some of the building location circular descriptors, using a circular earth mover's distance measurement of a distance between the image circular descriptor and the building location circular descriptor, and selecting one of the at least some building location circular descriptors that has a smallest measured distance to the image angular detector as the determined one building location circular descriptor.
A16. The computer-implemented method of any one of clauses A01-A15 further comprising obtaining a first enumerated group of ranges of angles, obtaining a second enumerated group of ranges of distances, and generating each of the building location circular descriptors by encoding information in that building location circular descriptor about some of the first latent space features by, for each of the at least some points of the structural elements that are visible from the building location of that building location circular descriptor, encoding information in that building location circular descriptor for one of 360 horizontal degrees from that building location to that point that includes one of the ranges of angles from the first enumerated group and one of the ranges of distances from the second enumerated group.
A17. The computer-implemented method of any one of clauses A01-A16 further comprising determining the position of the panorama image in the room by supplying, to a refinement neural network, the panorama image and building location with which the determined one building location circular descriptor is associated, and receiving an adjusted position that is based on that building location and is adjusted to reflect the visual information of the panorama image.
A18. The computer-implemented method of any one of clauses A01-A17 wherein the associating of the panorama image with the determined position and orientation further includes, by the computing device:
generating, for each of multiple building location circular descriptors associated with one of multiple building locations in the room, additional visual information for that building location circular descriptor that represents a view from the building location with which that building location circular descriptor is associated and that includes at least some of the second latent space features that are visible at the specified angular directions for that building location circular descriptor; and
determining an acquisition location of an additional image captured in the room by comparing an additional image circular descriptor generated for the additional image to the multiple building location circular descriptors, including using the generated additional visual information for the multiple building location circular descriptors.
A19. The computer-implemented method of clause A18 further comprising generating a graph having multiple nodes and with at least one node representing each of multiple rooms of the building, associating the multiple building location circular descriptors with one of the multiple nodes that represents the room, and further associating, after determining the position of the panorama image, the panorama image with the one node that represents the room.
A20. The computer-implemented method of any one of clauses A01-A19 wherein the comparing of the image circular descriptor to the building location circular descriptors includes using machine learning to identify the determined one building location circular descriptor as being most similar to the image circular descriptor.
A21. A computer-implemented method comprising multiple steps to perform automated operations that implement described techniques substantially as disclosed herein.
B01. A non-transitory computer-readable medium having stored executable software instructions and/or other stored contents that cause one or more computing systems to perform automated operations that implement the method of any of clauses A01-A21.
B02. A non-transitory computer-readable medium having stored executable software instructions and/or other stored contents that cause one or more computing systems to perform automated operations that implement described techniques substantially as disclosed herein.
B03. A non-transitory computer-readable medium having stored contents that cause one or more computing devices to perform automated operations including at least:
obtaining, by the one or more computing devices, and for an image captured in an area associated with a building and including visual information about at least some structural elements of the building, an image circular descriptor for the image that includes information identifying features associated with the at least some structural elements at specified directions within the visual information;
obtaining, by the one or more computing devices, building location circular descriptors each associated with a building location and including angular information about features associated with points of structural elements of the building at specified angular directions from the associated building location;
comparing, by the one or more computing devices, the image circular descriptor to the building location circular descriptors to determine one of the building location circular descriptors that has angular information best matching the information included in the image circular descriptor;
associating, by the one or more computing devices, the image with a determined position for the building that is based on the associated building location for the determined one building location circular descriptor; and
providing, by the one or more computing devices, information for the image about the determined position for the building.
B04. The non-transitory computer-readable medium of clause B03 wherein the image is a panorama image with 360 degrees horizontally of visual information, wherein the obtaining of the image circular descriptor includes generating the image circular descriptor by the one or more computing devices via analysis of the image by a trained neural network, and wherein the providing of the information about the determined position for the image includes presenting a floor plan for the building that includes a visual indication of the determined position for the image.
B05. The non-transitory computer-readable medium of any one of clauses B03-B04 wherein the area associated with the building includes at least one of multiple rooms of the building, and wherein the structural elements of the building include multiple of a door or a window or an inter-wall border.
B06. The non-transitory computer-readable medium of any one of clauses B03-B05 wherein the area associated with the building includes at least one external area proximate to the building, and wherein the structural elements of the building include multiple of a door or a window or an inter-wall border.
B07. The non-transitory computer-readable medium of any one of clauses B03-B06 wherein the visual information for the image has less than 360 horizontal degrees of coverage, wherein the determined one additional angular descriptor is for a panorama image that is taken at the determined position and that has 360 horizontal degrees of coverage, and wherein the comparing of the angular descriptor for the image to the additional angular descriptors includes matching the angular description for the image to a subset of the determined one additional angular descriptor for the panorama image.
C01. One or more computing systems comprising one or more hardware processors and one or more memories with stored instructions that, when executed by at least one of the one or more hardware processors, cause the one or more computing systems to perform automated operations that implement the method of any of clauses A01-A21.
C02. One or more computing systems comprising one or more hardware processors and one or more memories with stored instructions that, when executed by at least one of the one or more hardware processors, cause the one or more computing systems to perform automated operations that implement described techniques substantially as disclosed herein.
C03. A system comprising:
one or more hardware processors of one or more computing devices; and
one or more memories with stored instructions that, when executed by at least one of the one or more hardware processors, cause at least one of the one or more computing devices to perform automated operations including at least:
C04. The system of clause C03 wherein the recorded information includes a panorama image with visual information, wherein the structural elements include wall elements having at least one of a door or a window or an inter-wall border, and wherein the providing of the information about the determined position in the room includes presenting a floor plan for the building that includes the area, wherein the presented floor plan includes a visual indication of the determined position in the area.
C05. The system of any one of clauses C03-C04 wherein the area of the building is one of multiple rooms of the building.
C06. The system of any one of clauses C03-C05 wherein the area of the building is an external area adjacent to the building.
D01. A computer program adapted to perform the method of any of clauses A01-A21 when the computer program is run on a computer.
Aspects of the present disclosure are described herein with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the present disclosure. It will be appreciated that each block of the flowchart illustrations and/or block diagrams, and combinations of blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer readable program instructions. It will be further appreciated that in some implementations the functionality provided by the routines discussed above may be provided in alternative ways, such as being split among more routines or consolidated into fewer routines. Similarly, in some implementations illustrated routines may provide more or less functionality than is described, such as when other illustrated routines instead lack or include such functionality respectively, or when the amount of functionality that is provided is altered. In addition, while various operations may be illustrated as being performed in a particular manner (e.g., in serial or in parallel, or synchronous or asynchronous) and/or in a particular order, in other implementations the operations may be performed in other orders and in other manners. Any data structures discussed above may also be structured in different manners, such as by having a single data structure split into multiple data structures and/or by having multiple data structures consolidated into a single data structure. Similarly, in some implementations illustrated data structures may store more or less information than is described, such as when other illustrated data structures instead lack or include such information respectively, or when the amount or types of information that is stored is altered.
From the foregoing it will be appreciated that, although specific embodiments have been described herein for purposes of illustration, various modifications may be made without deviating from the spirit and scope of the invention. Accordingly, the invention is not limited except as by corresponding claims and the elements recited by those claims. In addition, while certain aspects of the invention may be presented in certain claim forms at certain times, the inventors contemplate the various aspects of the invention in any available claim form. For example, while only some aspects of the invention may be recited as being embodied in a computer-readable medium at particular times, other aspects may likewise be so embodied.
This application claims the benefit of U.S. Provisional Patent Application No. 63/270,794, filed Oct. 22, 2021 and entitled “Automated Analysis Of Visual Data Of Images To Determine The Images' Acquisition Locations On Building Floor Plans”; and of U.S. Provisional Patent Application No. 63/279,247, filed Nov. 15, 2021 and entitled “Automated Analysis Of Visual Data Of Images To Determine The Images' Acquisition Locations On Building Floor Plans”; each of which is hereby incorporated by reference in its entirety.
| Number | Date | Country | |
|---|---|---|---|
| 63270794 | Oct 2021 | US | |
| 63279247 | Nov 2021 | US |
