Systems and methods for estimating depth using ad hoc stereo array cameras

Abstract

Systems and methods for stereo imaging with camera arrays in accordance with embodiments of the invention are disclosed. In one embodiment, a method of generating depth information for an object using two or more array cameras that each include a plurality of imagers includes obtaining a first set of image data captured from a first set of viewpoints, identifying an object in the first set of image data, determining a first depth measurement, determining whether the first depth measurement is above a threshold, and when the depth is above the threshold: obtaining a second set of image data of the same scene from a second set of viewpoints located known distances from one viewpoint in the first set of viewpoints, identifying the object in the second set of image data, and determining a second depth measurement using the first set of image data and the second set of image data.

Description

FIELD OF THE INVENTION

The present invention relates to stereo imaging with camera arrays and how this extends the core capabilities of a single monolithic light field camera array.

BACKGROUND

An image captured by a digital camera provides some sense of the location of objects in a scene and the location of one object relative to another. Without information in a third dimension (depth), it can be difficult to make strict conclusions about locations and linear measurements.

A legacy stereo camera is a type of camera with two or more lenses with a separate image sensor for each lens. This allows the camera to simulate human binocular vision and the ability to capture three-dimensional (stereo) images. A legacy stereo camera has some ability to determine depth of objects in its field of view when the baseline, or distance, between the two cameras is known.

An array camera includes a plurality of individual imagers (i.e., cameras) that can capture images of a scene where the image obtained by each imager is from a slightly different viewpoint. The diversity of information between viewpoints can be used to calculate depth information. The depth calculations in an array camera are more sophisticated than in a stereo camera because additional combinations of images, from different cameras in the array, can be compared and correlated to make the estimates more robust in the presence of noise and aliasing. An array camera system typically still has reduced precision in determining depth beyond a certain distance from the camera because the information used to calculate depth becomes less diverse in magnitude the farther an object is from the camera.

SUMMARY OF THE INVENTION

Systems and methods for stereo imaging with camera arrays in accordance with embodiments of the invention are disclosed. In one embodiment, a method of generating depth information for an object in a scene using two or more array cameras that each include a plurality of imagers where each imager captures an image of the scene includes obtaining a first set of image data including image data of a scene captured from a first set of different viewpoints, identifying an object of interest in the first set of image data, determining a first depth measurement for the object of interest using at least a portion of the first set of image data, determining whether the first depth measurement for the object of interest is above a predetermined threshold, and when the depth for the object of interest is above the predetermined threshold: obtaining a second set of image data including image data of the same scene from a second set of different viewpoints located known distances from at least one viewpoint in the first set of different viewpoints, identifying the object of interest in the second set of image data, and determining a second depth measurement for the object of interest using at least a portion of the first set of image data and at least a portion of the second set of image data.

In a further embodiment, obtaining a first set of image data including image data of a scene captured from a first set of different viewpoints includes capturing a first set of image data using a first plurality of imagers in a first array.

Another embodiment also includes determining image capture settings for active imagers in the first array.

In a still further embodiment, determining image capture settings for active imagers in the first array includes calibrating for nonlinearities in the lenses of a plurality of the first plurality of imagers.

In still another embodiment, one of the imagers is designated as a reference camera and captures image data from a reference viewpoint.

In a yet further embodiment, identifying an object of interest in the first set of image data includes generating a preview image, presenting the preview image via a screen, and capturing user input.

In yet another embodiment, identifying an object of interest in the first set of image data includes detecting motion in an area of the scene.

In a further embodiment again, identifying an object of interest in the first set of image data includes detecting an object in a designated region of interest.

In another embodiment again, identifying an object of interest in the first set of image data includes selecting one or more pixels and associating the one or more pixels with the object.

In a further additional embodiment, obtaining a second set of image data including image data of the same scene from a second set of different viewpoints located known distances from at least one viewpoint in the first set of different viewpoints includes capturing a second set of image data using a second plurality of imagers in a second array.

In another additional embodiment, the second array is the first array placed in a different location from the location in which the first array captured the first set of image data.

A still yet further embodiment also includes estimating the baseline distance between the two arrays by cross-correlating one or more sets of corresponding pixels from the first set of image data and the second set of image data.

In still yet another embodiment, determining a first depth measurement for the object of interest using at least a portion of the first set of image data includes determining the disparity between the location of a pixel in one image in the first set of image data and a corresponding pixel in a second image in the first set of image data.

A still further embodiment again also includes calculating a confidence measure for the depth of the object of interest.

A still another embodiment again also includes determining if the object of interest is visible in the second set of image data and identifying the object of interest in the second set of image data when the object of interest is visible in the second set of image data.

In a still further additional embodiment, determining a second depth measurement for the object of interest using at least a portion of the first set of image data and at least a portion of the second set of image data includes determining the disparity between a first pixel associated with the object of interest in at least one image in the first set of image data and a corresponding second pixel in at least one image in the second set of image data.

In still another additional embodiment, determining a second depth measurement for the object of interest using at least a portion of the first set of image data and at least a portion of the second set of image data includes calculating the disparity between the location of a pixel in one image in the first set of image data and a corresponding pixel in a second image in the second set of image data.

In a yet further embodiment again, determining a second depth measurement for the object of interest using at least a portion of the first set of image data and at least a portion of the second set of image data includes utilizing the first depth measurement for the object of interest.

In yet another embodiment again, a method for calculating the speed of an object in a scene using one or more array cameras that each include a plurality of imagers where each imager captures an image of the scene includes obtaining a first set of image data including image data of a scene captured from a first set of different viewpoints, identifying an object of interest in the first set of image data, determining a first depth measurement and a first angular measurement for the object of interest using at least a portion of the first set of image data, determining a first location of the object of interest using at least the first depth measurement and first angular measurement, obtaining a second set of image data including a second image data of a scene captured from a second set of different viewpoints at a time t after the first set of image data was captured, identifying the object of interest in the second set of image data, determining a second depth measurement and a second angular measurement for the object of interest using at least a portion of the second set of image data, determining a second location of the object of interest using at least the second depth measurement and second angular measurement, calculating a speed for the object of interest using at least the first location and the second location of the object of interest.

In a yet further additional embodiment, obtaining a first set of image data including image data of a scene captured from a first set of different viewpoints includes capturing a first set of image data using a first plurality of imagers in a first array, and obtaining a second set of image data including a second image data of a scene captured from a second set of different viewpoints at a time t after the first set of image data was captured includes capturing a second set of image data using a second plurality of imagers at a time t after the first set of image data was captured.

In yet another additional embodiment, the second plurality of imagers is the same as the first plurality of imagers and the second set of different viewpoints is the same as the first set of different viewpoints.

In a further additional embodiment again, the second plurality of imagers is selected from a second array that is different from the first array.

Another additional embodiment again also includes determining image capture settings for active imagers in the first array.

In a still yet further embodiment again, identifying an object of interest in the first set of image data includes generating a preview image, presenting the preview image via a screen, and capturing user input.

In still yet another embodiment again, identifying an object of interest in the first set of image data includes detecting motion in an area of the scene.

In a still yet further additional embodiment, identifying an object of interest in the first set of image data includes detecting an object in a designated region of interest.

In still yet another additional embodiment, identifying an object of interest in the first set of image data includes selecting one or more pixels and associating the one or more pixels with the object.

In a yet further additional embodiment again, capturing a first set of image data using a first plurality of imagers in a first array includes capturing a first set of image data at a first location, capturing a second set of image data using a second plurality of imagers at a time t after the first set of image data was captured includes capturing a second set of image data in a second location at a time t after the first set of image data was captured, and determining a second location of the object of interest using at least the second depth measurement and second angular measurement includes determining the change from the first location to the second location.

In yet another additional embodiment again, capturing a first set of image data using a first plurality of imagers in a first array includes determining a first orientation of the first array, capturing a second set of image data using a second plurality of imagers at a time t after the first set of image data was captured includes determining a second orientation of the first array at a time t after the first set of image data was captured, and determining a second location of the object of interest using at least the second depth measurement and second angular measurement includes determining the change from the first orientation to the second orientation.

A still yet further additional embodiment again also includes calculating a confidence measure of the speed of the object of interest.

In still yet another additional embodiment again, determining a second depth measurement and a second angular measurement for the object of interest using at least a portion of the second set of image data includes determining a second depth measurement for the object of interest using at least a portion of the first set of image data and at least a portion of the second set of image data.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 is a conceptual illustration of an array camera architecture that can be used in a variety of array camera configurations in accordance with embodiments of the invention.

FIG. 2 is an illustration of parallax effects on the images captured in a two camera system.

FIG. 3A is a graph illustrating object distance versus observed disparity.

FIG. 3B is a graph illustrating object distance versus depth error.

FIG. 4 illustrates a process for measuring depth using stereo array cameras in accordance with embodiments of the invention.

FIG. 5 illustrates a process for reusing depth information in accordance with embodiments of the invention.

FIG. 6 illustrates a process for measuring speed using an array camera in accordance with embodiments of the invention.

DETAILED DISCLOSURE OF THE INVENTION

Turning now to the drawings, systems and methods for measuring distance and speed in accordance with embodiments of the invention are illustrated. In many embodiments of the invention, one or more array cameras are utilized to capture image data of a scene from slightly different viewpoints. The diversity of information acquired from different viewpoints can be used to calculate depth of objects in the scene and a depth map that includes a set of depth information for a scene. In many embodiments of the invention, a stereo array camera composed of two array cameras provides depth information that can be used to calculate depth with greater accuracy at distances farther from the camera than can a single array camera. In other embodiments, a stereo array camera is composed of an array camera and a legacy camera (i.e. having a single lens and single image sensor/imager).

Array cameras including camera modules that can be utilized to capture image data from different viewpoints are disclosed in U.S. patent application Ser. No. 12/935,504, entitled “Capturing and Processing of Images using Monolithic Camera Array with Heteregeneous Images”, filed May 20, 2009, the disclosure of which is incorporated by reference herein in its entirety. Array cameras offer a number of advantages and features over legacy cameras. An array camera typically contains two or more imagers (which can be referred to as cameras), each of which receives light through a separate lens system. The imagers operate to capture image data of a scene from slightly different viewpoints. Array cameras have a variety of applications, including capturing image data from multiple viewpoints that can be used in super-resolution processing and depth calculation.

Depth of an object of interest can be calculated by observing the disparity, or difference, in the location of corresponding pixels making up the object (pixels that capture the same content in a scene) in the images from two different cameras. As will be discussed further below, the baseline distance between the cameras, back focal length of the cameras, and disparity are factors in determining depth. The accuracy of a depth measurement is reduced with further distance from the camera because the disparity between the cameras of an image for a given object at that distance reduces with increasing distance. Eventually, the disparity gets smaller than a certain threshold sub-pixel amount for the given pixels size, and the resolution of depth measurement thus becomes more coarsely quantized with greater depth. Because increasing the baseline increases disparity, the accuracy of depth measurement can also be increased accordingly with the baseline. In many embodiments of the invention, a stereo array camera configuration includes two array cameras mounted a fixed distance apart. The fixed distance is greater than the distance between the cameras in a single array and the stereo array camera can therefore provide greater accuracy when making depth estimates than a single array.

The depth and two-dimensional location of an object of interest (such as from an image captured by an array camera) can be used to locate the object in three-dimensional space. Given two sets of three-dimensional coordinates of an object and the time elapsed, the object's speed and direction can be calculated to within a certain accuracy depending on the accuracy of the depth estimates in the two measurements. In several embodiments of the invention, speed and direction are calculated for an object from depth and two-dimensional location information determined using an array camera or stereo array camera. Array camera architectures that can be utilized for depth and speed measurements in accordance with embodiments of the invention are discussed below.

Array Camera Architecture

An array camera architecture that can be used in a variety of array camera configurations in accordance with embodiments of the invention is illustrated in FIG. 1. The array camera 100 includes an imager array 102, which is connected to a processor 108. Imagers 106 in the array 102 are evenly spaced in a 5×5 square. In other embodiments, imagers may have different spacing or can be arranged in other orientations in the array. The processor 104 is hardware, software, firmware, or a combination thereof that controls various operating parameters of the imager array 102. The processor 104 can also function to process the images received from imager array 102 to produce a synthesized higher resolution image using super-resolution processes, or transfer the images to other hardware, software, firmware or a combination thereof to process the images. The system can also include memory 108 in communication with the processor 104 for storing images. Architectures for imager arrays that can be utilized in accordance with embodiments of the invention include those disclosed in U.S. patent application Ser. No. 13/106,797, entitled “Architectures for System on Chip Array Cameras” to Pain et al., and U.S. patent application Ser. No. 12/935,504 entitled “Capturing and Processing of Images using Monolithic Camera Array with Heterogeneous Imagers” to Venkataraman et al., the disclosures of which are incorporated herein by reference in their entirety.

Although a specific architecture is illustrated in FIG. 1, any of a variety of architectures including any M×N array of cameras that enable the capture of low resolution images and application of super-resolution processes to produce a synthesized high resolution image can be utilized in accordance with embodiments of the invention.

Stereo Array Cameras

In many embodiments of the invention, two array cameras mounted a fixed distance apart form a pair of stereo array cameras. In other embodiments, an array camera and a legacy camera form a stereo array camera. Each array camera can be of an architecture as described above with respect to FIG. 1. In several embodiments, the array cameras have a similar configuration in features such as, but not limited to, number of cameras, array dimensions, camera resolution, and/or arrangement of color filters. In some embodiments, the active imagers (imagers used to capture image data) in the first and/or second array camera have color filters or other mechanisms such that they are configured to capture the same spectral band of light. The spectral band can be (but is not limited to) red, blue, green, infrared, or extended color. Extended color is a band that includes at least a portion of at the band of wavelengths of least two colors. Systems and methods for capturing and utilizing extended color are described in U.S. Patent Application No. 61/798,602, filed Mar. 15, 2013, entitled “Extended Color Processing on Pelican Array Cameras” and U.S. patent application Ser. No. 14/145,734, filed Dec. 31, 2013, entitled “Extended Color Processing on Pelican Array Cameras,” the disclosures of which are hereby incorporated by reference in their entirety. In other embodiments, the array cameras have a different configuration in one or more features as appropriate to a specific application. In many embodiments, the distance between the two arrays (or between the array and legacy camera) is known with reference to one or more locations on each array. In addition, the locations of each camera within each array are known. Therefore, the baseline (distance between any two cameras) between any camera in the first array and any camera in the second array (or legacy camera) is known or can be determined.

Distance Measurement Using Array Cameras

Images of a scene captured by different cameras in an array camera have differences due to the different points of view resulting from the different locations of the cameras, an effect known as parallax. These differences, referred to as disparity, provide information that can be used to measure depth of objects within a scene. Systems and methods for detecting disparity and calculating depth maps for an image are discussed in U.S. Patent Application Ser. No. 61/691,666 entitled “Systems and Methods for Parallax Detection and Correction in Images Captured Using Array Cameras” to Venkataraman et al., filed Aug. 21, 2012, the disclosure of which is incorporated by reference herein in its entirety.

Parallax in a two camera system is illustrated in FIG. 2. The two cameras 200, 202, include a lens stack 204 and a focal plane 206. Each camera has a back focal length f, and the two cameras are separated by the baseline distance of 2 h. The field of view of both cameras encompasses a scene including a foreground object 208 and a background object 210. The disparity introduced by the different fields of view of the two cameras 200, 202, is equal to the difference in location of the foreground object 208 between its location in the image captured by the first camera (represented as an offset of the point on the focal plane of the first camera 200 relative to its optical axis 212 shown as −u_L) and its location in the image captured by the second camera (represented as an offset of the point on the focal plane of the second camera 202 relative to its optical axis 214 is shown as u_R).

U.S. Patent Application Ser. No. 61/691,666 incorporated above discusses depth measurement using the following relationship between disparity and depth with respect to FIG. 2:

$\begin{array}{cc}{\Delta }_{\mathrm{parallax}}=u_R-u_L=\frac{2\mathrm{hf}}{z_o}& \left(1\right)\end{array}$

From the above equation and figure, it can be seen that disparity between images captured by the cameras is along a vector in the direction of the baseline of the two cameras, which can be referred to as the epipolar line between the two cameras. Furthermore, the magnitude of the disparity is directly proportional to the baseline separation of the two cameras and the back focal length of the cameras and is inversely proportional to the distance from the camera to an object appearing in the scene. The distance (or depth) from the two cameras to the foreground object can be obtained by determining the disparity of the foreground object in the two captured images.

One method of determining depth of a pixel or object using images captured by an array camera involves selecting an initial hypothesized depth or distance for a selected pixel from an image captured from a reference viewpoint/camera, and searching pixel locations in other images along the epipolar line between the reference viewpoint/camera and the camera capturing each of the other images for similar/matching pixels. This process is discussed in the patent incorporated by reference above, and can be modified to utilize two array cameras set farther apart than the cameras in a single array to determine depth to a higher precision as will be discussed further below.

Techniques such as those disclosed in the patent application incorporated above are typically used to generate a depth map from a reference viewpoint. The reference viewpoint can be from the viewpoint of one of the cameras in a camera array. Alternatively, the reference viewpoint can be an arbitrary virtual viewpoint. A depth map indicates the distance of the surfaces of scene objects from a reference viewpoint. Although a process for calculating depth using disparity is discussed above, any of a variety of techniques for calculating depth can be utilized in accordance with embodiments of the invention. Processes for depth measurement using stereo array cameras are discussed below.

Enhanced Distance Measurement Using Stereo Array Cameras

The closer that an object is to an array camera, the larger the disparity that will be observed in the object's location in different images captured by different cameras in the array. A representative graph of object distance with observed disparity is illustrated in FIG. 3A. It can be seen in the graph that as the object distance approaches zero (i.e., comes closer to the camera), the disparity increases dramatically. Conversely, as the object distance increases, the disparity decreases. It can also be seen that the rate of change in disparity decreases as object distance increases. A representative graph of object distance with depth error is illustrated in FIG. 3B. The graph assumes a 4×4 array where the baseline between any two adjacent cameras is 2.3 mm, the pixel size is 1.75 μm, and the focal length is about 2 mm. The depth error is calculated as the percentage of depth resolution over object distance, where depth resolution indicates the resolution of the depth (the distance by which two objects should be separated for the array camera to distinguish between them as two separate depths) at a given object distance. It can be seen that depth error increases with object distance.

The further a camera is from the reference viewpoint the larger the disparity that will be observed. Typically larger shifts enable depth to be determined with greater precision. Increasing the baseline (distance between cameras) increases the observed disparity accordingly. Therefore, using a camera that captures an image from a reference viewpoint and the cameras that are furthest from that camera to determine depth information can improve precision.

In many embodiments of the invention, two array cameras are set apart at a known distance in a stereo array camera configuration and image data from the two array cameras are used to generate depth information for an object observed by the cameras. In other embodiments, a stereo array camera includes an array camera and a legacy camera located a known distance from each other. A process for measuring depth using stereo array cameras in accordance with embodiments of the invention is illustrated in FIG. 4. The process includes determining (310) image capture settings for active cameras in a first array and a second array. Image capture settings can include calibration for nonlinearities or nonconformities in the lenses (e.g., by incorporating scene-independent geometric shifts as appropriate).

A first set of image data is captured (320) using active cameras in the first array. Typically, each camera collects image data that can be used to form an image from the point of view of the camera. In array cameras, often one camera is designated a reference camera and the image data captured by that camera is referred to as being captured from a reference viewpoint. In many embodiments of the invention, image data that is captured includes image data from a reference camera. In several embodiments, the active imagers capturing the image data are configured with color filters or other mechanisms to limit the spectral band of light captured. The spectral band can be (but is not limited to) red, blue, green, infrared, or extended color. Extended color is a band that includes at least a portion of at the band of wavelengths of least two colors. Systems and methods for capturing and utilizing extended color are disclosed in U.S. Patent Application No. 61/798,602 and Ser. No. 14/145,734 incorporated by reference above.

An object of interest is identified (325) in the first set of image data. The identification can be based upon a variety of techniques that include, but are not limited to: user input (e.g., selection on a screen), motion activation, shape recognition, and region(s) of interest. The identification can be made in an image generated from the first set of image data from the cameras in the first array. For example, the object of interest can be indicated in a preview image generated from the first set of image data or in a reference image from a reference viewpoint that corresponds to a reference camera in the first array. The identification can include selection of a pixel or set of pixels within the image associated with the object.

Using the first set of image data, a depth is determined (330) for the object. Techniques for determining the depth of the object can include those disclosed in U.S. Patent Application Ser. No. 61/691,666 incorporated by reference and discussed further above. The effects of noise can be reduced by binning or averaging corresponding pixels across images captured by different cameras utilizing techniques such as, but not limited to, those disclosed in U.S. Patent Application Ser. No. 61/783,441, filed Mar. 14, 2013, entitled “Systems and Methods for Reducing Motion Blur in Images or Video in Ultra Low Light with Array Cameras” to Molina and P.C.T. patent application Ser. No. 14/025,100, filed Mar. 12, 2014, entitled “Systems and Methods for Reducing Motion Blur in Images or Video in Ultra Low Light with Array Cameras” to Molina, the disclosures of which are hereby incorporated in their entirety. In several embodiments of the invention, intermediate images can be formed with pixel values in locations in each image where the pixel values are binned or averaged from corresponding pixels in different images. The intermediate images, which have noise components “averaged out” can then be used in depth calculation.

If the disparity of the object is above a predetermined threshold (340), i.e. is within a predetermined distance from the first array, the depth calculated above (330) is accepted as the depth of the object (350). A confidence measure can be given that is based on factors such as lens calibration and/or pixel resolution (the width that a pixel represents based on distance from the camera). The confidence measure can also incorporate information from a confidence map that indicates the reliability of depth measurements for specific pixels as disclosed in U.S. Patent Application Ser. No. 61/691,666 incorporated by reference above.

If the disparity of the object is below the predetermined threshold (340), then the depth measurement of the object is refined using a second set of image data from camera(s) in a second array. In some embodiments, the second array is instead a legacy camera. As discussed further above, a longer baseline between cameras can provide increased precision, because of increased disparity, out to further distances in measuring depth.

A second set of image data is captured (355) using at least one camera in the second array (or legacy camera). The object of interest is identified (370) in the second set of image data based upon a variety of techniques that can include those discussed above with respect to identifying the object in the first set of image data or other tracking techniques known in the art. If the system does not assume that the object of interest is visible to the second array, it can first determine (360) if the object is visible to at least one camera in the second array. Visibility can be determined, for example, by searching for similar pixels as discussed with respect to FIG. 9 in U.S. Patent Application Ser. No. 61/691,666 incorporated by reference above.

A depth measurement is performed (380) on the object using at least a portion of the first set of image data and at least a portion of the second set of image data. The measurement can include determining the disparity between pixel(s) associated with the object of interest in images captured by one or more cameras in the first array and corresponding pixel(s) in images captured by one or more cameras in the second array. In some embodiments, the second array is instead a legacy camera that captures a single image. The single image can similar be used as a second set of image data to determine disparity so long as pixel correspondences can be found between pixels in the first set of image data and the second set of image data.

Although specific processes are described above for obtaining depth measurements using multiple array cameras, any of a variety of combinations of two or more array cameras can be utilized to obtain depth measurements based upon the disparity observed between image data captured by cameras within the two array cameras can be utilized as appropriate to specific applications in accordance with embodiments of the invention.

A stereo array camera configuration can be formed in an ad hoc manner using one array camera and changing the position of the array camera. In many embodiments of the invention, an ad hoc stereo array camera includes an array camera capturing an image of a scene in one position, moving the array camera to a second position, and capturing a second image with the array camera in the second position. The two images captured in this way can form an ad hoc stereo pair of images. By correlating the features from the two images with each other and internal sensors such as a gyroscope and/or accelerometer in combination with the matched features, the camera extrinsics (such as camera center of projection and camera viewing direction) can be determined.

Unified Parallax Computation

A stereo array camera provides additional optimization possibilities in computing parallax disparities as compared to a single array camera. Parallax calculations can be performed using processes such as those disclosed in U.S. Provisional Patent Application Ser. No. 61/691,666 incorporated by reference above. As discussed above with respect to certain embodiments of the invention, parallax calculations can be performed to compute depths using the cameras in a first array in the stereo array camera. In many embodiments, information calculated using the first array can be used to accelerate calculation of depths with the second array in the stereo array camera. For example, in many processes for calculating depth, images are sampled for similar pixels to determine disparity as discussed in U.S. Provisional Patent Application Ser. No. 61/691,666. When pixels and/or objects have a depth that was already calculated by a first array, the search for similar pixels in the second array can use the depth information for the same pixel/object as a starting point and/or to limit the search to the “expected” portions of the image as predicted by the existing depth information. In several embodiments, the pixel/object can be correspondingly identified in images captured by the second array such that the existing depths can be applied to the proper pixel/object, even when the corresponding pixel/object is not in the same location within the image(s). In many embodiments, correspondence of pixels/objects is not necessarily determined for part or all of an image, but the depths of each pixel in the first image are used for calculating the depth of the pixel in the same location in the second image.

A process for reusing depth information in accordance with embodiments of the invention is illustrated in FIG. 5. The process includes determining (390) if depth was calculated for a pixel by a first array. If depth was not calculated, a depth is calculated (392) for the pixel using image data from the second array. If depth was calculated for the pixel, a depth is calculated (394) using image data from the second array taking into consideration the depth information from the first array, such as by limiting and/or refining the search for similar pixels as discussed above.

High Resolution Image Synthesis

The image data in low resolution images captured by a array camera can be used to synthesize a high resolution image using super-resolution processes such as those described in U.S. patent application Ser. No. 12/967,807 entitled “Systems and Methods for Synthesizing High Resolution Images Using Super-Resolution Processes” to Lelescu et al. The disclosure of U.S. patent application Ser. No. 12/967,807 is hereby incorporated by reference in its entirety. A super-resolution (SR) process can be utilized to synthesize a higher resolution (HR) 2D image or a stereo pair of higher resolution 2D images from the lower resolution (LR) images captured by an array camera. The terms high or higher resolution (HR) and low or lower resolution (LR) are used here in a relative sense and not to indicate the specific resolutions of the images captured by the array camera.

A stereo array camera configuration can also be used to create a HR image by using the cameras from both arrays. While the relatively large baseline between the two stereo array cameras would result in relatively larger occlusion zones (where parallax effects block some content that is captured in one camera from being captured in another camera), in other visible areas the cameras from the two arrays would enhance the final achieved solution. Preferably, each array camera is complete in its spectral sampling and utilizes a π color filter pattern so that the image that is synthesized using the cameras in one array is devoid of parallax artifacts in occlusion zones. In several embodiments, color filters in individual cameras can be used to pattern the camera module with π filter groups as further discussed in U.S. Provisional Patent Application No. 61/641,165 entitled “Camera Modules Patterned with pi Filter Groups”, to Nisenzon et al. filed May 1, 2012, the disclosure of which is incorporated by reference herein in its entirety.

High resolution (HR) images can be used to enhance depth measurement using stereo (two or more) array cameras in processes such as those described further above. In several embodiments of the invention, HR images are generated from image data captured by cameras in stereo array cameras. Each HR image can be generated using images captured by cameras in one array or images captured by cameras in both arrays. The HR images can then be used as image data in processes for generating depth measurement such as those described above. Measurement can be more robust using HR images because it is typically less sensitive to noise. Creating high resolution depth maps in accordance with embodiments of the invention is discussed below.

High Resolution Depth Map

The image data captured by a stereo array camera can be used to generate a high resolution depth map whose accuracy is determined by the baseline separation between the two arrays rather than the baselines of the individual cameras within either array. Depth maps can be generated by any of a variety of processes including those disclosed in U.S. Provisional Patent Application Ser. No. 61/691,666 incorporated by reference above. As discussed further above, the accuracy of depth measurement by an array camera is reduced at further distances from the camera. By using images captured by cameras in one array in a stereo array configuration with images captured by cameras in a second array, the baseline between the two cameras is significantly increased over the baseline between two cameras in a single array.

Auto Calibration of Stereo Array Cameras

A legacy stereo camera typically relies on a very accurate calibration between the two cameras to achieve the stereo effect. However, if the two cameras go out of alignment (e.g., by being dropped) the baseline between the two cameras becomes unknown. Without knowing the baseline, the ability to generate stereo imagery from the camera system is lost because the measured disparities cannot be converted into accurate estimates of depth.

With array cameras arranged in a stereo configuration in accordance with embodiments of the invention, each array individually can generate depth information for objects in a scene. By cross-correlating the pixels of the two array cameras or the depths calculated by the two array cameras, the baseline between the two array cameras can be estimated. This approach to estimating the baseline typically only works well when there are objects visible to both camera arrays whose depths can be calculated reasonably accurately using each camera array independently. If only objects at infinity are visible to both camera arrays, auto calibration as described here may not work. The depths calculated by a single array camera often will have some degree of error due to noise, nonlinearities or manufacturing defects in the lenses of the cameras, and/or other factors. The error can manifest in statistical variations in the depths calculated by the array camera. By correlating the depths calculated by one array in a stereo array camera with the depths calculated by the second array and/or depths calculated using images from one array together with images from the second array, an estimate can be made of the most likely baseline between the two array cameras in the stereo array.

Using the calculated baseline, the stereo array camera can calculate (or recalculate) depth to a higher precision for any object that is visible to both cameras in the array, such as by the processes outlined further above.

Near-Field and Far-Field Stereo

With a legacy stereo camera, an object is typically captured in stereo only if it is within the field of view of both (left and right) cameras. However, as the object comes closer to the stereo camera, it will eventually move out of the field of view of one of the cameras while still remaining in the field of view of the other camera. At this point, the stereo effect is lost because only one camera can “see” the object.

A stereo array camera in accordance with embodiments of the invention can generate both near-field and far-field stereo. As an object comes closer and moves out of the field of view of one array camera in a stereo configuration while staying within the field of view of the other array camera, it will still be captured in stereo. The cameras in the second array, which still “sees” the object, can be used to synthesize one or more virtual viewpoints (e.g., a left eye and right eye view). Good stereo acuity can be expected because the object will be close enough that the depth resolution will be high (i.e., precision of depth measurement). Processes for generating virtual viewpoints for stereo vision in accordance with embodiments of the invention are disclosed in U.S. Provisional Patent Application Ser. No. 61/780,906 entitled “Systems and Methods for Parallax Detection and Correction in Images Captured Using Array Cameras” to Venkataraman et al., filed Mar. 13, 2013, the disclosure of which is hereby incorporated by reference in its entirety.

Time elapsed between two images captured by a camera can be utilized with location information to provide a speed measurement. Speed measurement using array cameras in accordance with embodiments of the invention is discussed below.

Speed Measurement Using Array Cameras

Motion of an object across the field of view of a digital camera can generally be translated into an angular measurement (or angular velocity with elapsed time information) if the pixel size and back focal length are known, within the tolerance of one pixel and the corresponding angular measure of one pixel. At any given distance d from the camera, the angular measure of one pixel uniquely corresponds to a linear measure. Therefore, given a starting and ending location of an object in two dimensional images captured by a digital camera and the starting and ending distance of the object from the camera, the relative starting and ending locations of the object can be determined in three dimensional space. Provided the time elapsed between the images, the speed (or velocity) of the object can also be calculated. Given one start location and one end location, this can be represented as a linear velocity. Given multiple locations over time, the distance between each pair of consecutive locations (i.e. segment) can be determined and the distances of the segments combined to give a total distance. Additionally, a total average speed can be found by dividing the total distance over the time elapsed or by averaging the speed in each segment (distance divided by time elapsed in that segment) over the total time elapsed.

Conventional digital cameras typically capture two dimensional images without the capability of depth/distance measurement and are thus limited to angular measurement of motion. As discussed further above, array cameras can be used to determine depth by observing the disparity between multiple images that are captured by different cameras in the array. Formulas and techniques for determining distance relative to pixel disparity as in U.S. Patent Application Ser. No. 61/691,666 incorporated by reference above can also be used to determine the linear measure that the width of one pixel corresponds to at a given distance from the camera. In addition, one can calculate the time elapsed between the starting and ending frames simply by counting the number of frames between them and observing the frame rate of video capture of the camera.

In many embodiments of the invention, depth information for an object is combined with an angular measure of the object's position to provide a three-dimensional location for the object. In various embodiments of the invention, depth can be calculated using a single array camera or two array cameras in a stereo configuration as discussed further above. The three-dimension location of an object in two or more images can be used to calculate a speed and direction of the object. A process for measuring speed using an array camera in accordance with embodiments of the invention is illustrated in FIG. 6. The process includes determining (410) image capture settings for active cameras in an array camera. Image capture settings can include calibration for nonlinearities or nonconformities in the lenses (e.g., by incorporating scene-independent geometric shifts as appropriate). For example, due to physical characteristics of a lens stack, a pixel that is closer to the center line of a camera may differ in the area of a scene captured by the pixel from a pixel that is farther from the center line.

A first set of image data is captured (420) using active cameras in the array camera. Typically, each camera collects image data that can be used to form an image from the point of view of the camera. In array cameras, often one camera is designated a reference camera and the image data captured by that camera is referred to as being captured from a reference viewpoint. In many embodiments of the invention, depth measurements are made with respect to the viewpoint of the reference camera using at least one other camera (alternate view cameras) within the array.

An object of interest is identified (430) in the first set of image data. The identification can be based upon a variety of techniques that include, but are not limited to: user input (e.g., selection on a screen), motion activation, shape recognition, and region(s) of interest. The identification can be made in an image generated from the first set of image data from the cameras in the first array. For example, the object of interest can be indicated in a preview image generated from the first set of image data or in a reference image from a reference viewpoint that corresponds to a reference camera in the first array. The identification can include selection of a pixel or set of pixels within the image associated with the object.

Using the first set of image data, a first depth measure and a first location are determined (440) for the object. Techniques for determining the depth of the object can include those disclosed in U.S. Patent Application Ser. No. 61/691,666 incorporated by reference and discussed further above. Depth can be calculated using a single array camera or two array cameras in a stereo configuration as discussed further above. Using the two-dimensional location of the object in an image (e.g., a reference image) an angular measure can be determined for the location of the object with respect to the camera. Combining the angular measure with the depth measure gives a three-dimensional location of the object with respect to the camera. Any of a variety of coordinate systems can be utilized in accordance with embodiments of the invention to represent the calculated location of the object. In several embodiments of the invention, the centerline of a camera is treated as the origin.

At some time t after the capture of the first set of image data, a second set of image data is captured (450) using the cameras in the array. In many embodiments of the invention, the same set of cameras utilized to capture the first set of image data are used to capture the second set of image data. In other embodiments, a second set with a different combination of cameras is used to capture the second set of image data.

The object of interest is identified (460) in the second set of image data. Identification can be based upon a variety of techniques that can include those discussed above with respect to identifying the object in the first set of image data or other tracking techniques known in the art.

Using the second set of image data, a second depth measure and a second location are determined for the object (470). Depth can be calculated using techniques discussed further above using a single array camera or two array cameras in a stereo configuration. Location can be calculated using techniques discussed further above and can incorporate known information about the location of the second camera in relation to the first camera (e.g., removing parallax effects).

In different scenarios, an array camera used to capture sets of image data for speed measurement may be stationary (e.g., tripod mounted) or may be in motion (e.g., handheld or panning across a scene). It can also include an array camera using multiple image captures from slightly different points of view to get the advantage of a larger baseline and a more accurate depth. In several embodiments of the invention, an array camera is assumed to be stationary and need not compensate for motion of the camera. In other embodiments of the invention, an array camera includes sensors that collect camera motion information (480) on up to six degrees of movement of the camera, including motion along and rotation about three perpendicular axes. These sensors can include, but are not limited to, inertial sensors and MEMS gyroscopes. Camera motion information that is collected can be used to incorporate motion compensation when calculating the speed and/or direction of an object of interest (i.e., using the camera as a frame of reference). Motion compensation may be appropriate for functions such as stabilization (when there is jitter from slight movements of the camera such as by hand movement) or tracking an object (panning the camera to keep a moving object within the camera's field of view). In further embodiments of the invention, an array camera is configurable to switch between an assumption that it is stationary (no motion compensation) and that it is moving or moveable (apply motion compensation).

The speed of the object of interest is calculated (490) using the first location and second location of the object. The direction can also be calculated from the location information, as well as a vector representing the speed and direction of the object.

A confidence measure can be given that is based on factors such as lens calibration and/or pixel resolution (the width that a pixel represents based on distance from the camera). The confidence measure can also incorporate information from a confidence map that indicates the reliability of depth measurements for specific pixels as disclosed in U.S. Patent Application Ser. No. 61/691,666 incorporated by reference above.

Additionally, calculating speed in accordance with embodiments of the invention can involve calculating a refined depth measurement using two or more array cameras as discussed further above with respect to FIG. 4 or two separate image captures with the same array camera. The technique of using two different image captures of the same scene with the same array camera can also be extended to legacy cameras, wherein the legacy camera captures two images from different points of view, and depth and other measurements can be obtained through cross correlation of the images. In further embodiments of the invention, super-resolution processes can be utilized to perform measurements of angular position relative to the reference viewpoint at a higher precision. The super-resolution processes synthesize higher resolution images and the pixels in the synthesized higher resolution image enable measurement of the location of an object relative to an origin point within the image with higher precision. Accordingly, synthesizing a time sequence of higher resolution images using super-resolution processes can facilitate higher precision velocity measurements than measurements obtained using the lower resolution image data captured by the reference camera. Although a specific process for calculating speed is discussed above with respect to FIG. 6, any of a variety of processes for calculating speed can be utilized in accordance with embodiments of the invention.

Although the present invention has been described in certain specific aspects, many additional modifications and variations would be apparent to those skilled in the art. It is therefore to be understood that the present invention may be practiced otherwise than specifically described, including various changes in the implementation, without departing from the scope and spirit of the present invention. Thus, embodiments of the present invention should be considered in all respects as illustrative and not restrictive.

Claims

1. An array camera, comprising: a camera array comprising a plurality of cameras that capture images of a scene from different viewpoints;a gyroscope and accelerometers;a processor; andmemory in communication with the processor;wherein software directs the processor to: obtain a first set of images captured from different viewpoints at a first time using the camera array at a first location;obtain a second set of images captured from different viewpoints at a second time using the camera array at a second location distant from the first location;estimate extrinsic parameters for the camera array at the first time and at the second time based upon a combination of matching features in the first and second set of images and information from the gyroscope and accelerometers;select a reference viewpoint;determine depth estimates for pixel locations in an image from the reference viewpoint using the first set of images, wherein generating a depth estimate for a given pixel location in the image from the reference viewpoint comprises: identifying pixels in the first set of images that correspond to the given pixel location in the image from the reference viewpoint based upon expected disparity at a plurality of depths;comparing the similarity of the corresponding pixels identified at each of the plurality of depths; andselecting the depth from the plurality of depths at which the identified corresponding pixels have the highest degree of similarity as a depth estimate for the given pixel location in the image from the reference viewpoint;determine whether a depth estimate for pixel locations in an image from the reference viewpoint determined using the first set of images corresponds to an observed disparity below a threshold; andwhen the depth estimate corresponds to an observed disparity below the threshold, refining the depth estimate using images in the second set of images by: identifying pixels in the first and second sets of images that correspond to the given pixel location in the image from the reference viewpoint based upon expected disparity at a plurality of depths;comparing the similarity of the corresponding pixels form the first and second sets of images identified at each of the plurality of depths; andselecting the depth from the plurality of depths at which the identified corresponding pixels have the highest degree of similarity as a depth estimate for the given pixel location in the image from the reference viewpoint.

2. The array camera of claim 1, wherein software further directs the processor to select the plurality of depths at which pixels in images captured by the first array camera and by the second camera that correspond to the given pixel location in the image from the reference viewpoint are identified during refinement of the depth estimate based upon the depth estimate initially determined using the first set of images.

3. The array camera of claim 1, wherein software further directs the processor to: identify an object of interest in the first set of images;wherein determine depth estimates for pixel locations in an image from the reference viewpoint using the first set of images further comprises estimating a depth for the object of interest using the first set of images.

4. The array camera of claim 1, wherein software further directs the processor to calculate a confidence measure for the depth of an object of interest.

5. The array camera of claim 1, wherein software further directs the processor to generate a depth map using the depth estimates for pixel locations in an image from the reference viewpoint, where the depth map indicates distances of surfaces of scene objects from the reference viewpoint.

6. The array camera of claim 1, wherein the plurality of cameras in the camera array forms an M×N array of cameras.

7. The array camera of claim 1, wherein the camera array comprises cameras that image portions of the spectral band selected from the group consisting of red, blue, green, infrared, and extended color.

8. The array camera of claim 1, wherein the cameras in the camera array form a π filter group.

9. The array camera of claim 1, wherein: software further configures the processor to synthesize an image from the first set of images using the depth map;each image in the first set of images has a resolution; andthe image synthesized from the first set of images has a resolution that is greater than the resolution of each of the images in the first set of images.

10. The array camera of claim 1, wherein: software further configures the processor to synthesize an image from the first and second sets of images using the depth map and the extrinsic parameters of the camera array at the first and second times;each image in the first and second sets of images has a resolution; andthe image synthesized from the first set of images has a resolution that is greater than the resolution of each of the images in the first and second sets of images.

11. A method of estimating depth using an ad hoc stereo array camera, comprising: obtaining a first set of images captured from different viewpoints at a first time using a camera array;moving the camera array and obtaining a second set of images captured from different viewpoints at a second time using the camera array;obtaining gyroscope and accelerometer measurements between the first time and the second time;estimating extrinsic parameters for the camera array at the first time and at the second time based upon a combination of matching features in the first and second set of images and the gyroscope and accelerometers measurements;using software to direct a processor to: select a reference viewpoint;determine depth estimates for pixel locations in an image from the reference viewpoint using the first set of images, wherein generating a depth estimate for a given pixel location in the image from the reference viewpoint comprises: identifying pixels in the first set of images that correspond to the given pixel location in the image from the reference viewpoint based upon expected disparity at a plurality of depths;comparing the similarity of the corresponding pixels identified at each of the plurality of depths; andselecting the depth from the plurality of depths at which the identified corresponding pixels have the highest degree of similarity as a depth estimate for the given pixel location in the image from the reference viewpoint;determine whether a depth estimate for pixel locations in an image from the reference viewpoint determined using the first set of images corresponds to an observed disparity below a threshold; andwhen the depth estimate corresponds to an observed disparity below the threshold, refining the depth estimate using images in the second set of images by: identifying pixels in the first and second sets of images that correspond to the given pixel location in the image from the reference viewpoint based upon expected disparity at a plurality of depths;comparing the similarity of the corresponding pixels from the first and second sets of images identified at each of the plurality of depths; andselecting the depth from the plurality of depths at which the identified corresponding pixels have the highest degree of similarity as a depth estimate for the given pixel location in the image from the reference viewpoint.

12. The method of claim 11, further comprising using software to direct the processor to select the plurality of depths at which pixels in images captured by the first array camera and by the second camera that correspond to the given pixel location in the image from the reference viewpoint are identified during refinement of the depth estimatebased upon the depth estimate initially determined using the first set of images.

13. The method of claim 11, further comprising using software to direct the processor to: identify an object of interest in the first set of images;wherein determine depth estimates for pixel locations in an image from the reference viewpoint using the first set of images further comprises estimating a depth for the object of interest using the first set of images.

14. The method of claim 11, further comprising using software to direct the processor to calculate a confidence measure for the depth of an object of interest.

15. The method of claim 11, further comprising using software to direct the processor to generate a depth map using the depth estimates for pixel locations in an image from the reference viewpoint, where the depth map indicates distances of surfaces of scene objects from the reference viewpoint.

16. The method of claim 11, wherein the plurality of cameras in the camera array forms an M×N array of cameras.

17. The method of claim 11, wherein the camera array comprises cameras that image portions of the spectral band selected from the group consisting of red, blue, green, infrared, and extended color.

18. The method of claim 11, wherein the cameras in the camera array form a π filter group.

19. The method of claim 11, further comprising using software to direct the processor to synthesize an image from the first set of images using the depth map; wherein each image in the first set of images has a resolution; andwherein the image synthesized from the first set of images has a resolution that is greater than the resolution of each of the images in the first set of images.

20. The method of claim 11, further comprising using software to direct the processor to synthesize an image from the first and second sets of images using the depth map and the extrinsic parameters of the camera array at the first and second times; wherein each image in the first and second sets of images has a resolution; andwherein the image synthesized from the first and second sets of images has a resolution that is greater than the resolution of each of the images in the first and second sets of images.