WO2013177448A1

WO2013177448A1 - Systems and methods for feature tracking

Info

Publication number: WO2013177448A1
Application number: PCT/US2013/042504
Authority: WO
Inventors: Darren TURETZKY; Ryan ENGLE
Original assignee: 1-800 Contacts, Inc.
Priority date: 2012-05-23
Filing date: 2013-05-23
Publication date: 2013-11-28
Also published as: EP2852934A1; US20130314401A1; EP2852934A4; US10147233B2; US20130315487A1; WO2013177453A1; CA2874531C; WO2013177462A1; US9208608B2; US20130314411A1; EP2852934B1; US9378584B2; US9311746B2; WO2013177457A1; WO2013177459A1; AU2013266187A1; AU2013266187B2; US20130314410A1; EP2852935A1; US20150235428A1

Abstract

A computer-implemented method for processing, by a graphical processor unit (GPU), a plurality of images of a user. A plurality of features detected by the GPU in a first image of the plurality of images of the user is selected. Each selected feature includes one or more pixels. In a second image of the plurality of images of the user, a search is performed for the plurality of features selected in the first image. A variance is calculated, on the GPU, for each selected feature found in the second image. The variance indicates a degree to which a portion of the second image varies from a corresponding portion of the first image. The calculated variance is stored in a variance file.

Description

SYSTEMS AND METHODS FOR FEATURE TRACKING

RELATED APPLICATIONS

[0001] This application claims priority to U.S. Provisional Application No. 61/650,983, entitled SYSTEMS AND METHODS TO VIRTUALLY TRY-ON PRODUCTS, filed on May 23, 2012; U.S. Provisional Application No. 61/735,951 , entitled SYSTEMS AND METHODS TO VIRTUALLY TRY-ON PRODUCTS, filed on December 1 1 , 2012; and U.S. Application No. 13/775,764, filed on 25 February 2013, entitled SYSTEMS AND METHODS FOR FEATURE TRACKING, all of which are incorporated herein in their entirety by this reference.

BACKGROUND

[0002] The use of computer systems and computer-related technologies continues to increase at a rapid pace. This increased use of computer systems has influenced the advances made to computer-related technologies. Indeed, computer systems have increasingly become an integral part of the business world and the activities of individual consumers. Computers have opened up an entire industry of internet shopping. In many ways, online shopping has changed the way consumers purchase products. For example, a consumer may want to know what they will look like in and/or with a product. On the webpage of a certain product, a photograph of a model with the particular product may be shown. However, users may want to see more accurate depictions of themselves in relation to various products.

DISCLOSURE OF THE INVENTION

[0003] According to at least one embodiment, a computer-implemented method for processing, by a graphical processor unit (GPU), a plurality of images of a user. A plurality of features detected by the GPU in a first image of the plurality of images of the user may be selected. Each selected feature may include one or more pixels. In a second image of the plurality of images of the user, a search may be performed for the plurality of features selected in the first image. A variance may be calculated, on the GPU, for each selected feature found in the second image. The variance may indicate a degree to which a portion of the second image varies from a corresponding portion of the first image. The calculated variance may be stored in a variance file.

[0004] In one embodiment, one or more patches may be selected from among the selected features in the first image based on the calculated variance of each selected feature. Each patch may include a square area of pixels centered on one of the selected features of the first image. In some configurations, upon selecting the one or more patches in the first image, the first image may be removed from memory. In some embodiments, the one or more patches may be selected based on a predetermined threshold of calculated variance. In one embodiment, the one or more patches may be selected based on a predetermined number of patches. In some configurations, each variance may be divided into first and second elements. The first element may be stored in a first file and the second element may be stored in a second file.

[0005] In one embodiment, a cross-correlation algorithm may be performed on a GPU to determine how a first patch, selected among the one or more patches, is positioned in first and second sample images of the plurality of images of the user. In some embodiments, performing the cross-correlation algorithm on the GPU may include determining a pose of the user in the first and second sample images, performing a fast Fourier transform (FFT) on the first patch, and performing the FFT on the first and second sample images. The first sample image may be placed in the real element of a complex number and the second sample image may be placed in the imaginary element of the complex number. The FFT of the first patch may be stored in a third file. In some configurations, performing the cross-correlation algorithm on the GPU may include multiplying element-wise the FFT of the first patch by the FFT of the first and second sample images, calculating an inverse FFT of the multiplied FFTs, resulting in a first score for the first sample image and a second score for the second sample image, and normalizing the result of the cross-correlation by dividing both first and second scores by the calculated variance stored in the variance file.

[0006] In one embodiment, the cross-correlation of each selected patch may be performed on the GPU simultaneously. In some configurations, on the GPU, a second cross-correlation algorithm may be performed on third and fourth sample images of the plurality of images of the user using the FFT of the first patch stored in the third file to determine how the first patch is positioned in the third and fourth sample images. In some embodiments, a position of the selected feature of the first patch may be determined as a point in a virtual three-dimensional (3-D) space.

[0007] A computing device configured to process, by a graphical processor unit (GPU), a plurality of images of a user is also described. The device may include a processor and memory in electronic communication with the processor. The memory may store instructions that are executable by the GPU to select a plurality of features detected by the GPU in a first image of the plurality of images of the user. Each selected feature may include one or more pixels. In one embodiment, the instructions may be executable by the GPU to search, in a second image of the plurality of images of the user, for the plurality of features selected in the first image. In some configurations, the instructions may be executable by the GPU to calculate variance, on the GPU, for each selected feature found in the second image. The variance may indicate a degree to which a portion of the second image varies from a corresponding portion of the first image. In one embodiment, the instructions may be executable by the GPU to store the calculated variance in a variance file.

[0008] A computer-program product to process, by a graphical processor unit (GPU), a plurality of images of a user is also described. The computer-program product may include a non-transitory computer-readable medium that stores instruc- tions. The instructions may be executable by the GPU to select a plurality of features detected by the GPU in a first image of the plurality of images of the user. Each selected feature may include one or more pixels. In one embodiment, the instructions may be executable by the GPU to search, in a second image of the plurality of images of the user, for the plurality of features selected in the first image. In some configurations, the instructions may be executable by the GPU to calculate variance, on the GPU, for each selected feature found in the second image. The variance may indicate a degree to which a portion of the second image varies from a corresponding portion of the first image. In one embodiment, the instructions may be executable by the GPU to store the calculated variance in a variance file.

[0009] According to at least one embodiment, a computer-implemented method for processing, by a graphical processor unit (GPU), a plurality of images of a user. A plurality of features detected by the GPU in a first image of the plurality of images of the user may be selected. Each selected feature may include one or more pixels. In a second image of the plurality of images of the user, a search may be performed for the plurality of features selected in the first image. A variance may be calculated, on the GPU, for each selected feature found in the second image. The variance may indicate a degree to which a portion of the second image varies from a corresponding portion of the first image. The calculated variance may be stored in a variance file.

[0010] Features from any of the above-mentioned embodiments may be used in combination with one another in accordance with the general principles de- scribed herein. These and other embodiments, features, and advantages will be more fully understood upon reading the following detailed description in conjunction with the accompanying drawings and claims.

BRIEF DESCRIPTION OF THE DRAWINGS

[0011] The accompanying drawings illustrate a number of exemplary em- bodiments and are a part of the specification. Together with the following description, these drawings demonstrate and explain various principles of the instant disclosure.

[0012] FIG. 1 is a block diagram illustrating one embodiment of an environment in which the present systems and methods may be implemented;

[0013] FIG. 2 is a block diagram illustrating another embodiment of an environment in which the present systems and methods may be implemented;

[0014] FIG. 3 is a block diagram illustrating one example of a graphical processor unit (GPU);

[0015] FIG. 4 is a block diagram illustrating one example of a feature de- tection module;

[0016] FIG. 5 is a block diagram illustrating one example of a cross correlation module;

[0017] FIG. 6 is a diagram illustrating an example of a device for capturing an image of a user;

[0018] FIG. 7 illustrates an example arrangement of features detected in the depicted images of a user; [0019] FIG. 8 is a flow diagram illustrating one embodiment of a method for detecting features in images;

[0020] FIG. 9 is a flow diagram illustrating one embodiment of a method for performing cross-correlation algorithms on a GPU;

[0021] FIG. 10 is a flow diagram illustrating one embodiment of a method for performing a cross-correlation algorithm on two images simultaneously; and

[0022] FIG. 11 depicts a block diagram of a computer system suitable for implementing the present systems and methods.

[0023] While the embodiments described herein are susceptible to various modifications and alternative forms, specific embodiments have been shown by way of example in the drawings and will be described in detail herein. However, the exemplary embodiments described herein are not intended to be limited to the particular forms disclosed. Rather, the instant disclosure covers all modifications, equivalents, and alternatives falling within the scope of the appended claims. BEST MODE(S) FOR CARRYING OUT THE INVENTION

[0024] The systems and methods described herein relate to processing, by a graphical processor unit (GPU), a plurality of images of a user. Specifically, the systems and methods described herein relate to feature detection and normalized cross-correlation (i.e., template matching) of a set of images. Feature detection may include performing a "computer vision algorithm" to detect an "interesting" part of an image. Features may be used as a starting point in a computer vision algorithm. A desirable property of a feature detector may be repeatability, i.e., whether the same feature may be detected in one or more different images of the same scene. Typically, when working with feature detection and template matching of images, the image data may be copied from the GPU to the central processing unit (CPU), back to the GPU, and so forth. Copying image data back and forth between the GPU and CPU causes a bottleneck on the CPU, wasting valuable resources and computing cycles. On the CPU, feature detection of each image, as well as the cross-correlation of each interest point (i.e., corners, blobs, and/or points used in image analysis to detect distinguishable features of an image) on an image, may be performed one at a time, one after the other. However, because the results of one cross correlation do not depend on the results from another, feature detection and cross-correlation may be performed on the GPU simultaneously. Using a GPU instead of the CPU, the fast Fourier transform (FFT), element-wise multiplication, inverse FFT, and normalization may be performed simultaneously, all in parallel, for every interest point of a given image.

[0025] Additionally, in a typical implementation, the variance (degree of variation between pixels of two or more images) may not be calculated because of the computing costs involved calculating variance on the CPU. However, calculating variance on a GPU may be relatively fast compared to the CPU, and the results of the variance may be stored and used later in cross-correlation. However, rounding errors in the cross-correlation may be most affected when the variance of an interest point, or an extracted patch of the interest point, is low. Thus, in a GPU implementation, the most "trackable" features in an image may be those that have the highest variance.

[0026] FIG. 1 is a block diagram illustrating one embodiment of an environment 100 in which the present systems and methods may be implemented. In some embodiments, the systems and methods described herein may be performed on a single device (e.g., device 102). For example, a GPU 104 may be located on the device 102. Examples of devices 102 include mobile devices, smart phones, tablet computing devices, personal computing devices, computers, servers, etc.

[0027] In some configurations, a device 102 may include a GPU 104, a camera 106, and a display 108. In one example, the device 102 may be coupled to a database 1 10. In one embodiment, the database 1 10 may be internal to the device 102. In another embodiment, the database 1 10 may be external to the device 102. In some configurations, the database 1 10 may include variance data 1 12 and FFT patch data 1 14.

[0028] In one embodiment, the GPU 104 may enable feature detection and normalized cross-correlation to be performed in efficient, parallel operations. In some configurations, the GPU 104 may obtain multiple images of the user. For ex- ample, the GPU 104 may capture multiple images of a user via the camera 106. For instance, the GPU 104 may capture a video (e.g., a 5 second video) via the camera 106. In some configurations, the GPU 104 may use variance data 1 12 and FFT patch data 1 14 in relation to feature detection, cross-correlation, and 3-D modeling of a user. For example, the GPU may detect a degree of variation between a feature, or interest point, detected in a first image, and the same feature detected in a second image. The GPU 104 may store the detected variance in the variance data 1 12. In some configurations, the GPU 104 may generate a patch of an interest point. A patch may be a square set of pixels (e.g., l Opx by l Opx square of pixels) centered on the interest point. In some embodiments, the GPU 104 may perform an FFT algorithm on the patch. The GPU 104 may store the FFT of the patch in the FFT patch data 1 14. The GPU 104 may use, and reuse, the stored variance and FFT of the patch in relation to performing feature detection and feature tracking cross- correlation of one or more images.

[0029] FIG. 2 is a block diagram illustrating another embodiment of an environment 200 in which the present systems and methods may be implemented. In some embodiments, a device 102-a may communicate with a server 206 via a net- work 204. Example of networks 204 include, local area networks (LAN), wide area networks (WAN), virtual private networks (VPN), wireless networks (using 802.1 1 , for example), cellular networks (using 3G and/or LTE, for example), etc. In some configurations, the network 204 may include the internet. In some configurations, the device 102-a may be one example of the device 102 illustrated in FIG. 1 . For example, the device 102-a may include the camera 106, the display 108, and an application 202. It is noted that in some embodiments, the device 102-a may not include a GPU 104. In some embodiments, both a device 102-a and a server 206 may include a GPU 104 where at least a portion of the functions of the GPU 104 are performed separately and/or concurrently on both the device 102-a and the server 206.

[0030] In some embodiments, the server 206 may include the GPU 104 and may be coupled to the database 1 10. For example, the GPU 104 may access the variance data 1 12 and the FFT patch data 1 14 in the database 1 10 via the server 206. The database 1 10 may be internal or external to the server 206. In some embodiments, the database 1 10 may be accessible by the device 102-a and/or the server 206 over the network 204.

[0031] In some configurations, the application 202 may capture multiple images via the camera 106. For example, the application 202 may use the camera 106 to capture a video. Upon capturing the multiple images, the application 202 may process the multiple images to generate image data. In some embodiments, the application 202 may transmit one or more images to the server 206. Additionally or alternatively, the application 202 may transmit to the server 206 the image data or at least one file associated with the image data.

[0032] In some configurations, the GPU 104 may process multiple images of a user to detect features in an image, track the same features among multiple images, and determine a point in a virtual 3-D space corresponding to the tracked features. In some embodiments, the application 202 may process one or more image captured by the camera 106 in order to generate a 3-D model of a user.

[0033] FIG. 3 is a block diagram illustrating one example of a GPU 104-a. The GPU 104-a may be one example of the GPU 104 depicted in FIGS. 1 and/or 2. As depicted, the GPU 104-a may include a feature detection module 302 and a cross- correlation module 304.

[0034] In some configurations, the feature detection module 302 may examine a pixel of an image to determine whether the pixel includes a feature of interest. In some embodiments, the feature detection module 302 detects a face and/or head of a user in an image. In some embodiments, the feature detection module 302 detects features of the user's head and/or face. In some embodiments, the feature detection module 302 may detect an edge, corner, interest point, blob, and/or ridge in an image of a user. An edge may be points of an image where there is a boundary (or an edge) between two image regions, or a set of points in the image which have a relatively strong gradient magnitude. Corners and interest points may be used interchangeably. An interest point may refer to a point-like feature in an image, which has a local two dimensional structure. In some embodiments, the feature detection module 302 may search for relatively high levels of curvature in an image gradient to detect an interest point and/or corner (e.g., corner of an eye, corner of a mouth). Thus, the feature detection module 302 may detect in an image of a user's face the corners of the eyes, eye centers, pupils, eye brows, point of the nose, nostrils, cor- ners of the mouth, lips, center of the mouth, chin, ears, forehead, cheeks, and the like. A blob may include a complementary description of image structures in terms of regions, as opposed to corners that may be point-like in comparison. Thus, in some embodiments, the feature detection module 302 may detect a smooth, non- point-like area (i.e., blob) in an image. Additionally, or alternatively, in some embodiments, the feature detection module 302 may detect a ridge of points in the image. In some embodiments, the feature detection module 302 may extract a local image patch around a detected feature in order to track the feature in other images.

[0035] In some embodiments, the cross-correlation module 304 may process a feature detected by the feature detection module 302. The cross-correlation module 304 may track a feature detected by the feature detection module 302 and determine a position of a point in a virtual 3-D space corresponding to the tracked feature. Operations of the respective feature detection and cross-correlation modules 302 and 304 are discussed in further detail below.

[0036] FIG. 4 is a block diagram illustrating one example of a feature detection module 302-a. The feature detection module 302-a may be one example of the feature detection module 302 illustrated in FIG. 3. As depicted, the feature de- tection module 302 may include a selection module 402, a comparing module 404, a variance module 406, and a patch module 408.

[0037] In one embodiment, the selection module 402 may select a feature detected by the GPU 104 in a first image. The first image may be one of multiple images. The multiple images may include images of a user. Each detected feature may include one or more pixels. For example, the detected features may include edges, interest points, corners, blobs, ridges, and/or any other visual or pixel feature contained in an image (e.g., corners of the eyes, eye centers, point of the nose, corners of the mouth, etc.). In some configurations, the comparing module 404 may search a second image of the plurality of images of the user to find a match for the feature detected in the first image. For example, upon detecting the corner of a user's eye, the selection module 402 may select the detected corner of the user's eye as a feature to be tracked in subsequent images. The comparing module 404 may compare the detected corner of the user's eye in image 1 to the corresponding detected corner of the user's eye in image 2.

[0038] In some embodiments, the variance module 406 may calculate a variance for each selected feature found in the second image. The variance may indicate a degree a portion of the second image varies from a corresponding portion of the first image. For example, upon comparing the detected corner of the user's eye in image 1 to the corresponding detected corner of the user's eye in image 2, the variance module 406 may determine the degree of variation between the corner of the user's eye in image 2 and the corner of the user's eye in image 1. In some embodi- ments, the variance module 406 may write the calculated variance to a file and store the file in a storage medium. In some embodiments, the variance module 406 may store the variance in the variance data 1 12.

[0039] In some configurations, the variance module 406 may divide a calculated variance into first and second elements. The variance module 406 may store the first element in a first file and the second element in a second file within the variance data 1 12. In some embodiments, the variance module 406 divides the calculated variance based on a low order element of the variance and a high order element of the variance. For example, if the variance were calculated to be a decimal number equal to 123,456, then the variance module 406 may divide the hundreds element (i.e., low order element) from the thousands element (i.e., high order element). Storing the calculated variance as 3 -digit decimal numbers, the variance module 406 may store the thousands element " 123" in a first high order element file in the variance data 1 12, and may store the hundreds element "456" in a low order element file in the variance data 1 12. In some embodiments, the GPU 102 and the elements of the GPU 102 may access the high and low order element files stored in the variance data 1 12 in order to implement the calculated variance in one or more processes related to feature detection, detection tracking, cross-correlation, and generation of a 3-D model.

[0040] In one embodiment, the patch module 408 may select, or extract, from among the selected features in the first image one or more patches based on the calculated variance of each selected feature. Each extracted patch may include a square area of pixels centered on one of the selected features, or interest points, of the first image. Upon selecting and extracting the one or more patches in the first image, in some embodiments, the patch module 408 may remove the first image from memory. In some configurations, the patch module 408 may select and extract the one or more patches based on a predetermined threshold of calculated variance. Because a relatively low variance may cause a rounding error in the normalization of a cross-correlation process, in some embodiments, the patch module 408 may extract only those patches where the associated variance is above a certain level of variance. In some embodiments, the patch module 408 may select the one or more patches based on a predetermined number of patches.

[0041] FIG. 5 is a block diagram illustrating one example of a cross correlation module 304-a. The cross-correlation module 304-a may be one example of the cross-correlation module 304 illustrated in FIG. 3. As depicted, the cross- correlation module 304-a may include a pose detection module 502, a FFT module 504, a multiplication module 506, and a normalizing module 508.

[0042] In some configurations, the cross-correlation module 304-a may perform a cross-correlation algorithm to determine how a patch, selected among the one or more patches, is positioned in first and second sample images of the plurality of images of the user. In some embodiments, the cross-correlation module 304-a may determine a position of the selected feature of the first patch as a point in a vir- tual three-dimensional (3-D) space. In some configurations, as part of the process to perform the cross-correlation algorithm, the pose detection module 502 may determine a pose of the user in the first and second sample images. The FFT module 504 may perform a FFT on the first patch. In some embodiments, the FFT of the first patch may be written to a file and stored in the FFT patch data 1 14.

[0043] In some embodiments, the FFT module 504 may perform a FFT on the first and second sample images. The FFT module 504 may place the first sample image in a real element of a complex number (e.g., "a" in (a + bi)), and may place the second sample image in an imaginary element of the complex number (e.g., "b" in (a + bi)). In some configurations, the multiplication module 506 may multiply element-wise the FFT of the first patch by the FFT of the first and second sample images. The FFT module 504 may calculate an inverse FFT of the multiplied FFTs, resulting in a first score for the first sample image and a second score for the second sample image. In some embodiments, the normalization module 508 may normalize the result of the cross-correlation by dividing both first and second scores by the calculated variance stored in the variance file. For example, the normalization module 508 may access the high and low order element files stored in the variance data 1 12 to divide the first and second scores by the calculated variance. [0044] In some embodiments, the cross correlation module 304-a may perform the cross-correlation of every selected patch of a given image simultaneously. For example, the patch module 408 may extract 32 patches from image 1. The 32 patches may be cross-correlated simultaneously with image 2 and image 3 by placing image 2 and 3 in the elements of a complex number, performing the FFT of each patch, and normalizing the result by the stored variance. The FFTs of each patch may be stored by the patch module 408 in the FFT patch data 1 14 for subsequent cross-correlation operations. In some embodiments, the cross-correlation module 304-a may perform a second cross-correlation algorithm on images 4 and 5 using the same FFT patches stored in the FFT patch data 1 14.

[0045] FIG. 6 is a diagram 600 illustrating an example of a device 102-b for capturing an image 604 of a user 602. The device 102-b may be one example of the device 102 illustrated in FIGS. 1 and/or 2. As depicted, the device 102-b may include a camera 106-b, and display 108-b. The camera 106-b and display 108-b may be examples of the respective camera 106 and display 108 illustrated in FIGS. 1 and/or 2.

[0046] In one embodiment, the user may operate the device 102-b. For example, the application 202 may allow the user to interact with and/or operate the device 102-b. In one embodiment, the camera 106-a may allow the user to capture an image 604 of the user 602. The GPU 104 may perform feature detection and feature tracking in relation to the images of the user captured by the device 102-b. Additionally, or alternatively, the GPU 104 may perform a normalized cross-correlation algorithm on one or more of the images of the user to track one or more features detected in each image. The GPU 104 may determine a position of the selected feature as one or more points in a virtual three-dimensional (3-D) space to enable the generation of a 3-D model of the user.

[0047] FIG. 7 illustrates an example arrangement 700 of a feature 706 detected in the depicted images 702 and 704. As depicted, the example arrangement 700 may include a first image of the user 702 and a second image of the user 704. In some embodiments, the feature detection module 302 may detect a feature 708 on the first image of the user 702 (e.g., an eye of the user). The selection module 402 may select the feature 708. The patch module 408 may extract the selected feature 708 as a patch 706. Upon extracting the patch 706, in some embodiments, the patch module 408 may remove the image 702 from memory. The comparing module 404 may compare the patch 706 to the second image of the user 704. Upon finding a match between the patch 706 and the detected feature 710 of the of the second image 704, the variance module 406 may measure the degree of variation between the patch 706 and the detected feature 710. For example, the variation module 406 may detect that the detected feature 710 has shifted in one or more directions and/or changed shape (e.g., change in the shape of the eye due to rotation of the face) in relation to the patch 706. In some embodiments, the variance module 406 may store the vari- ance in the variance data 1 12. The stored variance may be used to normalize a cross-correlation of the patch 706 with one or more images. In some embodiments, the FFT module 504 may take the FFT of the patch 706 and store the FFT of the patch 706 in the FFT patch data 1 14. In some embodiments, the cross-correlation module 304 may use the stored FFT of the patch 706 in the performance of multiple cross-correlation algorithms.

[0048] FIG. 8 is a flow diagram illustrating one embodiment of a method 800 for detecting features in images. In some configurations, the method 800 may be implemented by the GPU 104 illustrated in FIGS. 1 , 2, and/or 3. In some configurations, the method 800 may be implemented by the application 202 illustrated in FIG. 2.

[0049] At block 802, a plurality of features detected by the GPU in a first image of the plurality of images of the user may be selected. Each selected feature may include one or more pixels. At block 804, in a second image of the plurality of images of the user, a search may be performed for the plurality of features selected in the first image.

[0050] At block 806, a variance may be calculated, on the GPU, for each selected feature found in the second image. At block 808, the variance may indicate a degree to which a portion of the second image varies from a corresponding portion of the first image. The calculated variance may be stored in a variance file.

[0051] FIG. 9 is a flow diagram illustrating one embodiment of a method

900 for performing cross-correlation algorithms on a GPU. In some configurations, the method 900 may be implemented by the GPU 104 illustrated in FIGS. 1 , 2, and/or 3. In some configurations, the method 900 may be implemented by the application 202 illustrated in FIG. 2.

[0052] At block 902, one or more patches may be selected from among selected features in a first image based on a calculated variance of each selected fea- ture. At block 904, a first cross-correlation algorithm may be performed on a GPU using a FFT of a first patch stored in a file to determine how the first patch is positioned in first and second sample images. At block 906, a second cross-correlation algorithm may be performed on the GPU on third and fourth sample images using the FFT of the first patch stored in the file to determine how the first patch is posi- tioned in third and fourth sample images.

[0053] FIG. 10 is a flow diagram illustrating one embodiment of a method 1000 for performing a cross-correlation algorithm on two images simultaneously. In some configurations, the method 1000 may be implemented by the GPU 104 illustrated in FIGS. 1 , 2, and/or 3. In some configurations, the method 1000 may be im- plemented by the application 202 illustrated in FIG. 2.

[0054] At block 1002, a pose of the user in the first and second sample images may be determined. At block 1004, a FFT may be performed on a patch extracted from an image. At block 1006, a FFT may be performed on the first and second sample images. The first sample image may be placed in a real element of a complex number and the second sample image may be placed in an imaginary element of the complex number.

[0055] At block 1008, the FFT of the first patch may be multiplied element- wise by the FFT of the first and second sample images. At block 1010, an inverse FFT of the multiplied FFTs may be calculated, resulting in a first score for the first sample image and a second score for the second sample image. At block 1012, the result of the cross-correlation may be normalized by dividing both first and second scores by the calculated variance stored in the variance file.

[0056] FIG. 11 depicts a block diagram of a computer system 1 100 suitable for implementing the present systems and methods. The depicted computer system 1 100 may be one example of a server 206 depicted in FIG 2. Alternatively, the system 1 100 may be one example of a device 102 depicted in FIGS. 1 , 2, and/or 6. Computer system 1 100 includes a bus 1 102 which interconnects major subsystems of computer system 1 100, such as a GPU 1 104, a system memory 1 106 (typically RAM, but which may also include ROM, flash RAM, or the like), an input/output controller 1 108, an external audio device, such as a speaker system 1 1 10 via an audio output interface 1 1 12, an external device, such as a display screen 1 1 14 via display adapter 1 1 16, serial ports 1 1 18 and mouse 1 146, a keyboard 1 122 (interfaced with a keyboard controller 1 124), multiple USB devices 1 126 (interfaced with a USB controller 1 128), a storage interface 1 130, a host bus adapter (HBA) interface card 1 136A operative to connect with a Fibre Channel network 1 138 , a host bus adapter (HBA) interface card 1 136B operative to connect to a SCSI bus 1 140, and an optical disk drive 1 142 operative to receive an optical disk 1 144. The GPU 1 104 may be one example of the GPU 104 depicted in FIGS. 1 , 2, and/or 3. Also included are a mouse 1 146 (or other point-and-click device, coupled to bus 1 102 via serial port 1 1 18), a modem 1 148 (coupled to bus 1 102 via serial port 1 120), and a network interface 1 150 (coupled directly to bus 1 102).

[0057] Bus 1 102 allows data communication between GPU 1 104 and system memory 1 106, which may include read-only memory (ROM) or flash memory (neither shown), and random access memory (RAM) (not shown), as previously noted. The RAM is generally the main memory into which the operating system and application programs are loaded. The ROM or flash memory can contain, among other code, the Basic Input-Output system (BIOS) which controls basic hardware operation such as the interaction with peripheral components or devices. For example, one or more instructions related to the operations of the GPU 1 104 to implement the present systems and methods may be stored within the system memory 1 106. Applications resident with computer system 1 100 are generally stored on and accessed via a non-transitory computer readable medium, such as a hard disk drive (e.g. , fixed disk 1 152), an optical drive (e.g. , optical drive 1 142), or other storage medium. Additionally, applications can be in the form of electronic signals modulated in accordance with the application and data communication technology when accessed via network modem 1 148 or interface 1 150.

[0058] Storage interface 1 130, as with the other storage interfaces of computer system 1 100, can connect to a standard computer readable medium for storage and/or retrieval of information, such as a fixed disk drive 1 152. Fixed disk drive 1 152 may be a part of computer system 1 100 or may be separate and accessed through other interface systems. Modem 1 148 may provide a direct connection to a remote server via a telephone link or to the Internet via an internet service provider (ISP). Network interface 1 150 may provide a direct connection to a remote server via a direct network link to the Internet via a POP (point of presence). Network interface 1 150 may provide such connection using wireless techniques, including digital cellular telephone connection, Cellular Digital Packet Data (CDPD) connection, digital satellite data connection or the like.

[0059] Many other devices or subsystems (not shown) may be connected in a similar manner (e.g. , document scanners, digital cameras and so on). Conversely, all of the devices shown in FIG. 1 1 need not be present to practice the present systems and methods. The devices and subsystems can be interconnected in different ways from that shown in FIG. 1 1 . The operation of at least some of the computer system 1 100 such as that shown in FIG. 1 1 is readily known in the art and is not dis- cussed in detail in this application. Code to implement the present disclosure can be stored in a non-transitory computer-readable medium such as one or more of system memory 1 106, fixed disk 1 152, or optical disk 1 144. The operating system provided on computer system 1 100 may be MS-DOS®, MS-WINDOWS®, OS/2®, UNIX®, Linux®, or another known operating system.

[0060] Moreover, regarding the signals described herein, those skilled in the art will recognize that a signal can be directly transmitted from a first block to a second block, or a signal can be modified (e.g. , amplified, attenuated, delayed, latched, buffered, inverted, filtered, or otherwise modified) between the blocks. Although the signals of the above described embodiment are characterized as trans- mitted from one block to the next, other embodiments of the present systems and methods may include modified signals in place of such directly transmitted signals as long as the informational and/or functional aspect of the signal is transmitted between blocks. To some extent, a signal input at a second block can be conceptualized as a second signal derived from a first signal output from a first block due to physical limitations of the circuitry involved (e.g., there will inevitably be some attenuation and delay). Therefore, as used herein, a second signal derived from a first signal includes the first signal or any modifications to the first signal, whether due to circuit limitations or due to passage through other circuit elements which do not change the informational and/or final functional aspect of the first signal.

[0061] While the foregoing disclosure sets forth various embodiments using specific block diagrams, flowcharts, and examples, each block diagram compo- nent, flowchart step, operation, and/or component described and/or illustrated herein may be implemented, individually and/or collectively, using a wide range of hardware, software, or firmware (or any combination thereof) configurations. In addition, any disclosure of components contained within other components should be considered exemplary in nature since many other architectures can be implemented to achieve the same functionality.

[0062] The process parameters and sequence of steps described and/or illustrated herein are given by way of example only and can be varied as desired. For example, while the steps illustrated and/or described herein may be shown or discussed in a particular order, these steps do not necessarily need to be performed in the order illustrated or discussed. The various exemplary methods described and/or illustrated herein may also omit one or more of the steps described or illustrated herein or include additional steps in addition to those disclosed.

[0063] Furthermore, while various embodiments have been described and/or illustrated herein in the context of fully functional computing systems, one or more of these exemplary embodiments may be distributed as a program product in a variety of forms, regardless of the particular type of computer-readable media used to actually carry out the distribution. The embodiments disclosed herein may also be implemented using software modules that perform certain tasks. These software modules may include script, batch, or other executable files that may be stored on a computer-readable storage medium or in a computing system. In some embodiments, these software modules may configure a computing system to perform one or more of the exemplary embodiments disclosed herein.

[0064] The foregoing description, for purpose of explanation, has been described with reference to specific embodiments. However, the illustrative discus- sions above are not intended to be exhaustive or to limit the invention to the precise forms disclosed. Many modifications and variations are possible in view of the above teachings. The embodiments were chosen and described in order to best ex- plain the principles of the present systems and methods and their practical applications, to thereby enable others skilled in the art to best utilize the present systems and methods and various embodiments with various modifications as may be suited to the particular use contemplated.

[0065] Unless otherwise noted, the terms "a" or "an," as used in the specification and claims, are to be construed as meaning "at least one of." In addition, for ease of use, the words "including" and "having," as used in the specification and claims, are interchangeable with and have the same meaning as the word "comprising." In addition, the term "based on" as used in the specification and the claims is to be con- strued as meaning "based at least upon."

Claims

What is claimed is:

1. A computer-implemented method for processing, by a graphical processor unit (GPU), a plurality of images of a user, the method comprising:

selecting a plurality of features detected by the GPU in a first image of the plurality of images of the user, wherein each feature comprises one or more pixels;

in a second image of the plurality of images of the user, searching for the plurality of features selected in the first image;

calculating, on the GPU, a variance for each selected feature found in the second image, wherein the variance indicates a degree a portion of the second image varies from a corresponding portion of the first image; and

storing the calculated variance in a variance file.

2. The method of claim 1 , further comprising:

selecting from among the selected features in the first image one or more

patches based on the calculated variance of each selected feature, wherein each patch comprises a square area of pixels centered on one of the selected features of the first image.

3. The method of claim 2, further comprising:

upon selecting the one or more patches in the first image, removing the first image from memory.

4. The method of claim 2, wherein selecting the one or more patches further comprises:

selecting the one or more patches based on a predetermined threshold of calculated variance.

5. The method of claim 2, wherein selecting the one or more patches further comprises:

selecting the one or more patches based on a predetermined number of patches.

6. The method of claim 1 , further comprising:

dividing each variance into first and second elements; and

storing the first element in a first file and the second element in a second file.

7. The method of claim 1 , further comprising:

performing on the GPU a cross-correlation algorithm to determine how a first patch, selected among the one or more patches, is positioned in first and second sample images of the plurality of images of the user.

8. The method of claim 7, wherein performing the cross-correlation algorithm on the GPU comprises:

determining a pose of the user in the first and second sample images;

performing a fast Fourier transform (FFT) on the first patch, wherein the FFT of the first patch is stored in a third file;

performing the FFT on the first and second sample images, placing the first sample image in a real element of a complex number and placing the second sample image in an imaginary element of the complex number; multiplying element-wise the FFT of the first patch by the FFT of the first and second sample images;

calculating an inverse FFT of the multiplied FFTs, resulting in a first score for the first sample image and a second score for the second sample image; and

normalizing the result of the cross-correlation by dividing both first and second scores by the calculated variance stored in the variance file.

9. The method of claim 8, further comprising:

performing on the GPU a second cross-correlation algorithm on third and fourth sample images of the plurality of images of the user using the FFT of the first patch stored in the third file to determine how the first patch is positioned in the third and fourth sample images.

10. The method of claim 7, wherein performing on the GPU the cross-correlation algorithm further comprises

performing on the GPU the cross-correlation of each selected patch simulta- neously.

1 1. The method of claim 1 , further comprising:

determining a position of the selected feature of the first patch as a point in a virtual three-dimensional (3-D) space.

12. A computing device configured to process, by a graphical processor unit (GPU), a plurality of images of a user, comprising:

the GPU;

memory in electronic communication with the GPU;

instructions stored in the memory, the instructions being executable by the

GPU to:

select a plurality of features detected by the GPU in a first image of the plurality of images of the user, wherein each feature comprises one or more pixels;

in a second image of the plurality of images of the user, search for the plurality of features selected in the first image;

calculate, on the GPU, a variance for each selected feature found in the second image, wherein the variance indicates a degree a portion of the second image varies from a corresponding portion of the first image; and

storing the calculated variance in a variance file.

13. The computing device of claim 12, wherein the instructions are executable by the GPU to:

select from among the selected features in the first image one or more patches based on the calculated variance of each selected feature, wherein each patch comprises a square area of pixels centered on one of the selected features of the first image;

upon selecting the one or more patches in the first image, remove the first image from memory.

14. The computing device of claim 13, wherein the instructions are executable by the GPU to:

select the one or more patches based on a predetermined threshold of calculated variance.

15. The computing device of claim 12, wherein the instructions are executable by the GPU to:

perform on the GPU a cross-correlation algorithm to determine how a first patch, selected among the one or more patches, is positioned in first and second sample images of the plurality of images of the user.

16. The computing device of claim 15 , wherein performing the cross-correlation algorithm on the GPU comprises instructions executable by the GPU to:

determine a pose of the user in the first and second sample images;

perform a fast Fourier transform (FFT) on the first patch, wherein the FFT of the first patch is stored in a third file;

perform the FFT on the first and second sample images, placing the first sample image in a real element of a complex number and placing the second sample image in an imaginary element of the complex number; multiply element-wise the FFT of the first patch by the FFT of the first and second sample images;

calculate an inverse FFT of the multiplied FFTs, resulting in a first score for the first sample image and a second score for the second sample image; and

normalize the result of the cross-correlation by dividing both first and second scores by the calculated variance stored in the variance file.

17. The computing device of claim 16, wherein the instructions are executable by the GPU to:

perform on the GPU a second cross-correlation algorithm on third and fourth sample images of the plurality of images of the user using the FFT of the first patch stored in the third file to determine how the first patch is positioned in the third and fourth sample images.

18. The computing device of claim 15, wherein the instructions are executable by the GPU to:

perform on the GPU the cross-correlation of each selected patch simultaneously.

19. A computer-program product for processing, by a graphical processor unit (GPU), a plurality of images of a user, the computer-program product comprising a non-transitory computer-readable medium storing instructions thereon, the instructions being executable by the GPU to:

store the calculated variance in a variance file.

20. The computer-program product of claim 19, wherein the instructions are executable by the GPU to:

perform on the GPU a first cross-correlation algorithm of each selected patch simultaneously to determine how each patch is positioned in first and second sample images of the plurality of images of the user; perform on the GPU a second cross-correlation algorithm of each selected patch simultaneously to determine how each patch is positioned in third and fourth sample images of the plurality of images of the user; and

normalize the results of the first and second cross-correlations by dividing the results by the calculated variance stored in the variance file.