The present invention relates to a providing apparatus and a providing method for performing processing relating to a virtual viewpoint image, and a computer readable storage medium.
A free viewpoint image (virtual viewpoint image) technique is an example of a technique for reproducing an image from a virtual camera disposed at any position in a three-dimensional space using multiple real camera images. With the virtual viewpoint image technique, an image taken from any virtual camera position is generated by estimating the three-dimensional shapes of objects. By transmitting model data (three-dimensional shapes and texture images) of the objects to a terminal of a user, a virtual viewpoint image corresponding to an interactive operation performed by the user can be generated. However, since the data amount of the model data of an object is large, the transmission of the model data puts pressure on the communication band. Japanese Patent No. 5563545 proposes a configuration in which the configuration density of the three-dimensional shapes is varied according to the amount of change in the shapes as a method for reducing the amount of data to be transmitted.
However, Japanese Patent No. 5563545 focuses only on the configuration density of shapes, and therefore there is a risk that information that is necessary for the user will be omitted. Accordingly, the means for reducing the amount of data disclosed in Japanese Patent No. 5563545 is not suitable for generating model data for generating a virtual viewpoint image.
An embodiment of the present invention discloses a providing apparatus and a providing method for providing three-dimensional geometric data, according to which the data amount of model data for regenerating a virtual viewpoint image can be efficiently reduced.
According to one aspect of the present invention, there is provided a providing apparatus configured to provide three-dimensional geometric data to be used to generate a virtual viewpoint image, the providing apparatus comprising: a hardware processor; and a memory storing one or more programs configured to be executed by the hardware processor, the one or more programs including instructions for: receiving a data request from a communication apparatus; deciding which of a plurality of pieces of three-dimensional geometric data including first three-dimensional geometric data and second three-dimensional geometric data with a different quality than the first three-dimensional geometric data is to be provided to the communication apparatus from which the received data request was transmitted; and providing the three-dimensional geometric data decided on from among the plurality of pieces of three-dimensional geometric data, to the communication apparatus as a response to the received data request.
According to another aspect of the present invention, there is provided a providing method for providing three-dimensional geometric data to be used to generate a virtual viewpoint image, the providing method comprising: receiving a data request from a communication apparatus; deciding which of a plurality of pieces of three-dimensional geometric data including first three-dimensional geometric data and second three-dimensional geometric data with a different quality than the first three-dimensional geometric data is to be provided to the communication apparatus from which the received data request was transmitted; and providing the three-dimensional geometric data decided on from among the plurality of pieces of three-dimensional geometric data, to the communication apparatus as a response to the received data request.
Furthermore, according to another aspect of the present invention, there is provided a non-transitory computer readable storage medium storing a program for causing a computer to execute a providing method for providing three-dimensional geometric data to be used to generate a virtual viewpoint image, the providing method comprising: receiving a data request from a communication apparatus; deciding which of a plurality of pieces of three-dimensional geometric data including first three-dimensional geometric data and second three-dimensional geometric data with a different quality than the first three-dimensional geometric data is to be provided to the communication apparatus from which the received data request was transmitted; and providing the three-dimensional geometric data decided on from among the plurality of pieces of three-dimensional geometric data, to the communication apparatus as a response to the received data request.
Further features of the present invention will become apparent from the following description of exemplary embodiments (with reference to the attached drawings).
Hereinafter, an embodiment of the present invention will be described with reference to the drawings. Note that the following embodiment does not limit the present invention, and not all combinations of features described in the present embodiment are essential in the solving means of the present invention. Note that identical configurations are denoted by identical reference signs in the description.
In the first embodiment, a method for transmitting model data that is needed when regenerating an interactive virtual viewpoint image in a user terminal will be described. Here, in order to simplify the description, a case is envisioned in which there is one server, which is a transmission source, and there are multiple clients, which are reception destinations. The location is set inside of a stadium in which a soccer game is held, an image processing apparatus, which is a server, is present in the stadium, and in a spectator seat, a user operates a display apparatus (a terminal such as a smartphone or a tablet), which is a client, to browse a virtual viewpoint image. Note that in the present embodiment, a virtual viewpoint image means an image viewed from a viewpoint that is set virtually. Terms such as “free viewpoint image” and “arbitrary viewpoint image” are also similar to “virtual viewpoint image”.
A LAN 108 is formed with wires or wirelessly, and is used for data transmission and reception between an image processing apparatus, a camera apparatus group, a display apparatus group, and an analysis apparatus. The camera apparatus group is constituted by multiple cameras 120. The cameras 120 are connected to an image processing apparatus 100 via the LAN 108, and based on control signals from the image processing apparatus 100, the cameras 120 start and stop image capture, change camera settings (shutter speed, focal distance, diaphragm value, etc.), and transfer image capture data. The display apparatus group is constituted by multiple user terminals 130 (smartphones, tablets, etc.). Each user terminal 130 is connected to the image processing apparatus 100 via the LAN 108, and receives model data that is needed for browsing a virtual viewpoint image from the image processing apparatus 100. The user terminal 130 uses the received model data to generate and display a virtual viewpoint image. Since the communication band of the LAN 108 is limited, the size of the model data that can be received by the user terminal 130 depends on the number of users. The analysis apparatus 140 analyzes the type of play of the object using the images from the cameras 120 and various types of information from sensors provided on the objects. Note that the analysis apparatus 140 is optional and is not an essential constituent component. Note that various constituent components other than those described above are present in the system configuration. For example, various devices may be connected via the Internet, a WAN, or the like instead of the LAN 108. Also, for example, the image processing system may be formed due to the image processing apparatus 100, the multiple cameras 120, and the analysis apparatus 140 being connected via the LAN 108, and the image processing system and the user terminal 130 may be connected through the Internet or the like.
In step S402 and onward, object geometric data indicating the three-dimensional shapes of specific objects such as players and texture data corresponding to the object geometric data is generated in multiple classes with different qualities. In step S402, the CPU 101 transmits a change in the camera settings for making light exposure during image capture suitable, and a signal for starting image capture to the cameras 120. In response to the signal for starting image capture, the cameras 120 start image capture and transfer the image data to the image processing apparatus 100 via the LAN 108. The image processing apparatus 100 receives the image data from the cameras 120 and deploys it to the main memory 102. The image processing apparatus 100 manages the image data as multi-viewpoint frames obtained by consolidating the image frames of the cameras for each identical time code. Also, at this time, the image processing apparatus 100 calculates the positions and orientations of the cameras using a method such as Structure from Motion and stores the positions and orientations of the cameras.
In step S403, the CPU 101 extracts the outlines of a group of objects serving as specific objects from the image data and generates the three-dimensional shapes and positions of the group of objects using a method such as Visual-hull. The outline extraction of the object group can be acquired by using an intermediate value filter in all frames of a camera image of one camera. Also, the three-dimensional shape is output as point cloud data or voxel data. This processing is carried out on multi-viewpoint frames of all time codes, and geometric point cloud data (geometric data expressed using a high-density point cloud) of all objects is generated in units of multi-viewpoint frames. The generated geometric point cloud data is stored in the storage unit 103.
In step S404, the CPU 101 thins out the geometric point cloud data generated in step S403, connects the remaining point clouds to form meshes constituting surfaces (triangular polygons), and thus generates meshes indicating the objects. A known technique can be applied in the mesh formation, and for example, a method such as Ball Pivoting can be used. The CPU 101 executes mesh formation processing on all of the geometric point cloud data generated in units of multi-viewpoint frames and stores the obtained data (low-density mesh data) in the storage unit 103. In step S405, the CPU 101 generates texture data (textures of objects) to be attached to the meshes generated in step S404. A known technique can be applied to the generation of the textures of the objects. The CPU 101 generates the textures for all of the mesh data generated in units of multi-viewpoint frames and stores the obtained data in the storage unit 103.
In step S406, the CPU 101 reads out the meshes and the textures generated from the multi-viewpoint frame (initial frame) corresponding to the first time code of the image from the storage unit 103 and expands them to the main memory 102. In step S407, the CPU 101 incorporates bones for controlling the orientations of the meshes in the meshes loaded in step S406. As shown in
In step S408, the CPU 101 estimates the joint positions of all of the objects using the camera images. A known technique can be applied in the estimation of the joint positions. For example, the joint positions (x(n,i,k,t), y(n,i,k,t)) in two dimensions in the images are acquired using machine learning. Here, 0≤x<horizontal width of image, 0≤y<vertical width of image, 0≤n<number of cameras, 0≤i<number of objects, 0≤k<number of joints, and 0≤t<number of frames. After the two-dimensional joint positions are acquired in the images of at least two or more cameras, joint positions (X(i,k,t), Y(i,k,t), Z(i,k,t)) in three dimensions are acquired using triangulation based on the positions of the cameras obtained in step S402. At this time, X, Y, and Z are coordinate values in a three-dimensional space, 0≤i<number of objects, 0≤k<number of joints, and 0≤t<number of frames. Accordingly, the movement paths of the joints of the objects, that is, the transitions of the orientations are acquired.
In step S409, the CPU 101 associates the joint positions of the bones generated in step S407 with the three-dimensional joint positions generated in step S408 in units of multi-viewpoint frames, and thus causes a transition in the orientations of (animates) the meshes and bones. Accordingly, with the meshes and textures, only the data of the multi-viewpoint frame at the start of the image need be prepared, and a small amount of animation data indicating the paths of the joint positions need only be added, and therefore the model data is substantially compressed in the time axis direction, and the amount of data can be significantly reduced.
In step S410, the CPU 101 holds the model data generated in steps S403 to S409 as the class structure indicated in Table 5a of
In step S411, if the analysis apparatus 140 is present, the CPU 101 associates the model data with play information (play content such as shooting, passing, clearance) of the objects, which is the analysis data obtained from the analysis apparatus 140. Accordingly, for example, it is possible to carry out processing and the like for extracting three-dimensional geometric data and texture data of a desired class during shooting. According to the above-described processing, the generation of model data having a class structure is complete.
The required classes are described in the game category attribute for each type of game serving as an attribute of the content, as shown in Table 6a of
Regarding the area attribute, as shown in Table 6c in
With the game attribute, as shown in Table 6d of
Note that in order to display the content selection window 901, the user terminal 130 acquires a list of pieces of content to be selected from the image processing apparatus 100. Each piece of content in the list corresponds to a group of multi-viewpoint frames that are temporally continuous. For example, based on the play content (analysis result of the analysis apparatus 140), one piece of content may be generated using a series of multi-viewpoint frames including time codes close to that at which the play occurred. For example, as described in step S411, the model data associated with each piece of play information may be used as one piece of content. Alternatively, for example, separate pieces of content may be generated for the multi-viewpoint frames of the first half of a match and the multi-viewpoint frames of the last half of the match. Note that the positions and orientations (directions) of the virtual camera that are automatically set based on the play content and the occurrence position of the play may be defined in each piece of content.
In step S801, upon receiving a request for a piece of content to be transmitted from the user terminal 130, the image processing apparatus 100 decides on the class to be transmitted through the processing of step S804 and onward, and transmits the model data for the decided class. First, in step S804, the CPU 101 of the image processing apparatus 100 acquires a vacant band of the communication line. In step S805, the CPU 101 sets the spec class of the model data based on the spec information received from the user terminal 130. For example, if the capabilities of the CPU and the GPU are low-end, model data of class 3 or class 2, which have high processing loads, cannot be processed, and therefore the spec class is set to class 1. If the display resolution is low, differences between classes are difficult to see, and therefore the spec class is set to class 2 or lower (i.e., class 1 or class 2). In step S806, the CPU 101 determines whether or not the spec class set in step S805 is class 1. If it is class 1, the processing advances to step S811, and if not, the processing advances to step S807.
In step S807, the CPU 101 generates model data for transmission using the attribute data generated in step S303. As shown in
In step S808, the CPU 101 determines whether or not the model data for transmission can be transmitted based on the vacant band of the communication line acquired in step S804 and the size of the model data for transmission generated in step S807. If it is determined that communication is possible (YES), the processing advances to step S814, and if it is determined that communication is not possible (NO), the processing advances to step S809. In step S809, the CPU 101 generates the model data for transmission with the required classes described in the attribute data reduced by one level. For example, the required class for area 0 in Table 6c of
In step S815, the user terminal 130 (CPU 131) receives the model data. In step S816, the CPU 131 uses the received model data to generate a virtual viewpoint image and regenerates the virtual viewpoint image on the display unit 135. Upon receiving the model data from the image processing apparatus 100, the CPU 131 executing the application in the user terminal 130 transitions to the virtual viewpoint window 903 shown in
In step S1003, the CPU 131 determines whether or not the model data of class 2 is present in the transmitted model data. If the result of the determination is YES, the processing advances to step S1004, and if the result of the determination is NO, the processing advances to step S1005. In step S1004, the CPU 131 uses the model data of class 2 to generate an image at the set position, direction, and angle of view of the virtual camera (image 2). In step S1005, the CPU 131 determines whether or not the model data of class 3 is present in the transmitted model data. If the result of the determination is YES, the processing advances to step S1006, and if the result of the determination is NO, the processing advances to step S1010. In step S1006, the CPU 131 uses the model data of class 3 to generate an image at the set position, direction, and angle of view of the virtual camera (image 3).
In step S1007, the CPU 131 determines whether or not there is a difference in the classes of the player models at continuous time codes (a previous time code and a current time code). This is a case in which, for example, the class of player 2 is 1 at time code 0, and the class of player 2 is 3 at time code 1. If there is a difference, the processing advances to step S1008, and if there is no difference, the processing advances to step S1009. In step S1008, the CPU 131 generates an object image by compositing (e.g., alpha blending) the image 1 and the image 2, or the image 1 and the image 3. This is done to prevent the image quality from changing rapidly when there is a difference in classes between time codes. On the other hand, in step S1009, the CPU 131 generates the object image by replacing the object region of image 1 with object image 2 or image 3, which have high image quality. In step S1010, the CPU 131 generates a background image by rendering the background model. In step S1011, the CPU 131 generates the virtual viewpoint image by compositing the object image and the background image. Note that in step S1007, if the class of the player 2 is 2 at time code 0 but the class is 1 at time code 1 (if the class decreases at continuous time codes), it is determined that there is no difference. This is because in this kind of case, the importance level of the object decreases, and thus there is no problem if the image quality rapidly changes. Note that in the description above, the image is composited if the image changes from class 1 to class 2 or class 3 and the image is not composited otherwise, but there is no limitation to this. For example, the object image of the unchanged class and the object image of the changed class may be composited when the class of the model data of the object changes.
As described above, with the image processing apparatus of the first embodiment, three-dimensional object data for each object is generated with multiple classes, and the importance levels of the objects is decided based on the characteristics of the contents, such as the analysis results of a game category or an actual game. Also, since the class of the model data constituting the model data for transmission is set according to the importance level, the model data for the virtual viewpoint image corresponding to the interactive operation can be generated suitably and transmitted efficiently.
In the second embodiment, a configuration in which the attribute data is corrected based on the degree of involvement in the game, the attention level of an object, and user preference will be described. Note that description of systems and processing that is redundant with the description of the first embodiment will be omitted.
In the first embodiment, the type of game, area, and occurring events such as an important play are used to determine the importance of the model data of the objects and decide the classes to be used. In the second embodiment, the required class is further changed using the attention level of a player who is an object (e.g., whether or not the player is famous), user preference (e.g., whether or not the user likes the player), and degree of involvement in the game (e.g., distance from the ball). In
In step S1202, the CPU 101 determines whether or not the size of the model data for transmission will increase after modification. If the result of the determination is YES, the processing advances to step S1203, and if the result of the determination is NO, the processing ends. In step S1203, the class is reduced according to the degree of involvement (distance from the ball) in Table 12a in order to reduce the size of the model data for transmission. For example, although the player 2 was at class 3 after step S1201, the distance between the player 2 and the ball increases, and therefore it is determined that the degree of involvement in the game of the player 2 is low, and the player 2 is lowered to class 2. For example, the CPU 101 decides on the model data (three-dimensional geometric data, texture data) to be provided to the display apparatus 130 such that the quality of the three-dimensional geometric data of a player (specific object) whose distance from the ball is less than a threshold value is higher than the quality of the three-dimensional geometric data of a player whose distance from the ball is greater than or equal to the threshold value, among the multiple players.
As described above, according to the second embodiment, when a class to be used is selected from model data of multiple classes generated for each object, attributes relating to the individual objects, such as the attention level of the object, user preference, and degree of involvement in the game, are considered. As a result, it is possible to generate more suitable model data for transmission, and it is possible to efficiently transmit the model data for the virtual viewpoint image corresponding to the interactive operation. Note that in the description above, the degree of involvement, the attention level, and the preference are illustrated as attributes relating to the individual objects, but there is no limitation thereto. Although both the attention level and the preference were considered in step S1201, it is also possible to use a configuration in which one of them is considered.
In the second embodiment, a configuration was described in which the required class is modified based on the attributes set for the individual objects. In the third embodiment, a configuration will be described in which the required classes are modified according to the position, orientation, and angle of view of the virtual camera to optimize the model data for transmission when the three-dimensional data is transmitted as a stream. Note that description of systems and processing that is redundant with the description of the first embodiment and the second embodiment will be omitted.
In stream transmission, the model data at time code M is transmitted from the image processing apparatus 100 and is received and regenerated by the display apparatus. Thereafter, the position and orientation of the virtual camera are fed back from the user terminal 130, and based on the fed-back position and orientation, the image processing apparatus 100 transmits model data for time code M+1. By sequentially repeating this processing, the interactive virtual viewpoint image can be regenerated by the display apparatus without waiting for all of the data to be received. At this time, a more optimal class can be selected with consideration given to the virtual camera position and orientation of the previous time code.
In step S1302, the CPU 101 determines whether or not the size of the model data for transmission will increase after modification. If the result of the determination is YES, the processing advances to step S1303, and if the result of the determination is NO, the processing ends. In step S1303, the CPU 101 lowers the class according to the movement speed of the virtual camera and the distance between the virtual camera and the players in order to reduce the size. Note that the movement speed of the virtual camera is calculated based on the amount of change in the position and orientation in the previous frame and the frame before that. For example, a case is considered in which the movement speed of the virtual camera is fast. In this case, although the player 2 had class 2 after step S1301, the distance from the virtual camera is large, and therefore the player 2 moves quite fast in the image. For this reason, it is determined that hardly any difference in image quality for each class will occur, and the classes are lowered to class 1.
As described above, according to the third embodiment, the classes of the objects are modified based on the position/orientation/angle of view of the virtual camera designated by the user terminal, and therefore suitable model data for transmission can be generated according to the state of the virtual camera. Note that in the above-described embodiments, description was given with a focus on an example in which model data of multiple classes having different data sizes are generated for each object. However, it is also possible to generate model data of multiple classes for only one or more specific objects among the multiple objects.
Embodiment(s) of the present invention can also be realized by a computer of a system or apparatus that reads out and executes computer executable instructions (e.g., one or more programs) recorded on a storage medium (which may also be referred to more fully as a ‘non-transitory computer-readable storage medium’) to perform the functions of one or more of the above-described embodiment(s) and/or that includes one or more circuits (e.g., application specific integrated circuit (ASIC)) for performing the functions of one or more of the above-described embodiment(s), and by a method performed by the computer of the system or apparatus by, for example, reading out and executing the computer executable instructions from the storage medium to perform the functions of one or more of the above-described embodiment(s) and/or controlling the one or more circuits to perform the functions of one or more of the above-described embodiment(s). The computer may comprise one or more processors (e.g., central processing unit (CPU), micro processing unit (MPU)) and may include a network of separate computers or separate processors to read out and execute the computer executable instructions. The computer executable instructions may be provided to the computer, for example, from a network or the storage medium. The storage medium may include, for example, one or more of a hard disk, a random-access memory (RAM), a read only memory (ROM), a storage of distributed computing systems, an optical disk (such as a compact disc (CD), digital versatile disc (DVD), or Blu-ray Disc (BD)™), a flash memory device, a memory card, and the like.
While the present invention has been described with reference to exemplary embodiments, it is to be understood that the invention is not limited to the disclosed exemplary embodiments. The scope of the following claims is to be accorded the broadest interpretation so as to encompass all such modifications and equivalent structures and functions.
This application claims the benefit of Japanese Patent Application No. 2017-179010, filed Sep. 19, 2017, which is hereby incorporated by reference herein in its entirety.
Number | Date | Country | Kind |
---|---|---|---|
2017-179010 | Sep 2017 | JP | national |
Number | Name | Date | Kind |
---|---|---|---|
20170142486 | Masuda | May 2017 | A1 |
20170322017 | Aoki | Nov 2017 | A1 |
20180041762 | Ikai | Feb 2018 | A1 |
Number | Date | Country |
---|---|---|
1809844 | Jul 2006 | CN |
101271591 | Sep 2008 | CN |
H09-282249 | Oct 1997 | JP |
H10-198823 | Jul 1998 | JP |
2004206338 | Jul 2004 | JP |
2013-183209 | Sep 2013 | JP |
2014-086012 | May 2014 | JP |
5563545 | Jul 2014 | JP |
2014126906 | Jul 2014 | JP |
2017139725 | Aug 2017 | JP |
Entry |
---|
Lamboray, E., Waschbüsch, M., Würnnlin, S., Pfister, H., & Gross, M. H. (2003). Dynamic point cloud compression for free viewpoint video. CS technical report, 430. (Year: 2003). |
Decision to Grant a Patent issued by the Japanese Patent Office dated Oct. 9, 2018 (date of drafting of Oct. 2, 2018) in corresponding Japanese Patent Application No. 2017-179010, with English translation. |
First Office Action issued by the State Intellectual Property Office of the People's Republic of China dated Jul. 30, 2019 in corresponding Chinese Patent Application No. 201811074027.X with English translation. |
Lamboray et al., “Dynamic Point Cloud Compression for Free Viewpoint Video,” ETH CS Technical Report #430, Dec. 10, 2003, pp. 1-7. |
Collet et al., “High-Quality Streamable Free-Viewpoint Video,” ACM Transactions on Graphics, vol. 34, No. 4, Article 69, Aug. 2015, pp. 69:1-69:13. |
Salvador et al., “From Silhouettes to 3D Points to Mesh, Towards Free Viewpoint Video,” 3DVP'I0, Oct. 29, 2010, pp. 19-24. |
Kajitani et al., “Point Cloud Streaming for 3D Avatar Communication,” I-Tech, Nov. 1, 2008, pp. 372-388. |
Extended European Search Report issued by the European Patent Office dated Mar. 21, 2019 in corresponding European Patent Application No. 18192401.0. |
Japanese Office Action corresponding to JP Patent Application No. 2017-179010 dated Jun. 15, 2018. |
Notification of Reason for Refusal issued by the Korean Intellectual Property Office dated Jan. 16, 2020 in corresponding Korean Patent Application No. 10-2018-0111179 with English translation. |
Number | Date | Country | |
---|---|---|---|
20190089942 A1 | Mar 2019 | US |