Gait recognition feature extraction method, gait recognition method, device and medium

Abstract

This application belongs to the field of video image processing technology, specifically relating to a gait recognition feature extraction method, gait recognition method, device, and medium. The method includes: S1, acquiring video image data of the target to be detected, wherein the video image data is encoded using the MPEG-4 video compression standard; S2, for P-frame images in the image group of the video image data, based on a preset residual image entropy threshold, selecting residual images whose residual image entropy is greater than the residual image entropy threshold as residual images to be optimized, and using the corresponding motion vectors as motion vectors to be optimized; S3, based on a preset segmentation threshold, segmenting the residual images to be optimized to obtain a gait contour map; S4, optimizing the motion vectors to be optimized based on the gait contour map to obtain optimized motion vectors. This method reduces the computational load and resource consumption in the gait recognition process, improves the speed and accuracy of gait recognition, and reduces storage overhead.

Description

Technical Field

[0001] This application belongs to the field of video image processing technology, specifically relating to a gait recognition feature extraction method based on video encoded data. Background Technology

[0002] Gait recognition is a technology that relies on the posture and body shape information of a person during walking to identify them. It has the advantage of being difficult to spoof at long distances and has broad application prospects in fields such as criminal investigation and security.

[0003] Existing gait recognition technologies typically input continuous gait contour image sequences into a neural network to identify the target object. However, in reality, there is a large amount of redundant information between frames in the original video sequence. The massive amount of video data increases storage overhead and makes it difficult for the neural network model to quickly extract effective information, significantly increasing the time cost during model training. In practical applications, this also leads to slower recognition speeds.

[0004] Existing methods that use motion vectors or residuals from video coding to represent optical flow motion avoid video decoding and optical flow calculation. However, because P-frames only retain the difference from the previous frame (I-frame or P-frame), they contain almost no useful information when the redundancy between adjacent frames is too high, resulting in highly sparsity data. Furthermore, motion vectors are usually calculated by block matching estimation algorithms, which are prone to errors due to environmental factors such as pixel changes, resulting in data containing a lot of noise. Directly using motion vectors or residuals to represent optical flow motion for gait recognition has low accuracy.

[0005] Existing methods for gesture recognition using motion vectors have achieved good results, accurately identifying stationary hand shapes as well as changes in moving gestures, i.e., macroscopic movements and main directions of movement. However, when applied to gait recognition, they cannot extract the features of gait contour changes over a continuous time period. In other words, such fine gait contours cannot be provided by motion vectors. Therefore, existing methods for gesture recognition using motion vectors have low recognition accuracy when used for gait recognition.

[0006] In summary, improving recognition speed and accuracy while reducing storage overhead are technical problems that urgently need to be solved by those skilled in the art. Summary of the Invention

[0007] (a) Technical problems to be solved

[0008] In view of the above-mentioned shortcomings and deficiencies of the prior art, this application provides a gait recognition feature extraction method, gait recognition method, device and medium.

[0009] (II) Technical Solution

[0010] To achieve the above objectives, this application adopts the following technical solution:

[0011] In a first aspect, embodiments of this application provide a gait recognition feature extraction method based on video encoded data, the method comprising the following steps:

[0012] S1. Acquire video image data of the target to be detected, wherein the video image data is encoded using the MPEG-4 video compression standard;

[0013] S2. For the P-frame images of the image group in the video image data, based on the preset residual image entropy threshold, select residual images whose residual image entropy is greater than the residual image entropy threshold as residual images to be optimized, and use the corresponding motion vectors as motion vectors to be optimized.

[0014] S3. Based on a preset segmentation threshold, a gait contour map is obtained by segmenting from the residual image to be optimized;

[0015] S4. Optimize the motion vector to be optimized based on the gait profile map to obtain the optimized motion vector.

[0016] Optionally, step S2 includes:

[0017] The residual images of the three channels in the P-frame image are converted into grayscale images using a weighted average method.

[0018] Calculate the image entropy of each grayscale residual image;

[0019] Grayscale residual images with residual image entropy greater than a preset residual image entropy threshold are selected as residual images to be optimized, and the corresponding motion vectors are selected as motion vectors to be optimized.

[0020] Optionally, the residual image of the three channels in the P-frame image is converted into a grayscale image and then converted to a size of 100*100 using the following formula:

[0021] Gray(x,y)=0.299Red(x,y)+0.587Green(x,y)+0.114Blue(x,y)

[0022] Where Gray is a grayscale image, Red, Green, and Blue are the residual images of the red, green, and blue channels, respectively, and x and y are the horizontal and vertical coordinates of the pixel.

[0023] Optionally, the method further includes the following steps after S2 and before S3:

[0024] The residual image to be optimized is subjected to bilateral filtering using the following formula.

[0025]

[0026] Where η is the normalization factor, Let be the filtered image, Δ(p) be the image before filtering, q be the nearest points of pixel p, Ω represent the spatial range, and w(x) and σ(x) are both Gaussian functions, with w(x) being the spatial kernel and σ(x) being the range kernel.

[0027] Optionally, step S3 includes: based on a preset segmentation threshold, segmenting the residual image to be optimized using the following segmentation function to obtain a gait contour map:

[0028]

[0029] Among them, B k (x,y) represents the gait profile segmented from the residual, Δ k (x,y) represents the pixel value at coordinates (x,y) in the residual map, and T represents the segmentation threshold.

[0030] Optionally, step S4 includes:

[0031] The gait profile is aligned with the motion vector to be optimized and expanded into a dual-channel model.

[0032] Based on the gait profile, the motion vector to be optimized is optimized according to the following formula to obtain the optimized motion vector:

[0033]

[0034] Where λ represents the optimization weight, mv k (x,y) represents the motion vector to be optimized. This represents the optimized motion vector.

[0035] Optionally, the preset residual image entropy threshold is set to 1.

[0036] Secondly, embodiments of this application provide a gait recognition method, the method comprising:

[0037] The optimized motion vector of the target to be detected is obtained by the gait recognition feature extraction method based on video encoded data as described in any of the first aspects above.

[0038] The optimized motion vector is input into a pre-established gait recognition model to obtain the identity recognition information of the target to be detected.

[0039] Thirdly, embodiments of this application provide an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, it implements the steps of the gait recognition feature extraction method based on video encoded data as described in any of the first aspects above.

[0040] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the gait recognition feature extraction method based on video encoded data as described in any of the first aspects above.

[0041] (III) Beneficial Effects

[0042] The beneficial effects of this application are as follows: This application proposes a gait recognition feature extraction method, gait recognition method, device, and medium. The method includes: S1, acquiring video image data of the target to be detected, wherein the video image data is encoded using the MPEG-4 video compression standard; S2, for P-frame images of image groups in the video image data, based on a preset residual image entropy threshold, selecting residual images whose residual image entropy is greater than the residual image entropy threshold as residual images to be optimized, and using the corresponding motion vectors as motion vectors to be optimized; S3, based on a preset segmentation threshold, segmenting the residual images to be optimized to obtain a gait contour map; S4, optimizing the motion vectors to be optimized based on the gait contour map to obtain optimized motion vectors. This method reduces the computational load and the occupancy rate of computing resources in the gait recognition process, improves the speed and accuracy of gait recognition, and reduces storage overhead. Attached Figure Description

[0043] This application is described with reference to the following figures:

[0044] Figure 1 This is a schematic diagram of the gait recognition feature extraction method based on video encoded data in one embodiment of this application;

[0045] Figure 2 This is a network model framework diagram for gait recognition in one embodiment of this application;

[0046] Figure 3 This is a schematic flowchart of a gait recognition method in another embodiment of this application;

[0047] Figure 4 This is a schematic diagram of the architecture of an electronic device in another embodiment of this application. Detailed Implementation

[0048] To better explain and facilitate understanding of the present invention, it is described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described below are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other; for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.

[0049] Before introducing the gait recognition feature extraction method based on video encoded data in this application, we will first explain the MPEG-4 video compression coding standard used in the video encoded data involved in this application.

[0050] The MPEG-4 video compression coding method divides video into I-frames (intra-frame compressed frames), P-frames (inter-frame predicted frames), and B-frames (bidirectional predicted frames). I-frames are keyframes in the video, usually RGB images containing complete information about the video frame. P-frames are inter-frame predicted frames, each predicted from the preceding P-frame or I-frame. B-frames are bidirectional predicted frames, predicted from the preceding I-frame or P-frame and the following P-frame. Both P-frames and B-frames are calculated using motion estimation and compensation methods. A complete compressed video is composed of stacked I-frames, P-frames, and B-frames.

[0051] Video compression coding algorithms primarily involve two parts: intra-frame predictive compression and inter-frame predictive compression. The compression algorithm first segments each frame, dividing it horizontally into several lines and then vertically into several blocks. These blocks are called macroblocks, typically using a 16×16 macroblock as the basic unit. Next, the video frames are grouped, calculating the similarity of macroblocks between adjacent frames. Based on a threshold, highly correlated frames are merged into a group. Each group retains only the complete image data of the first frame, called an I-frame. Other frames are calculated by referencing the previous frame and are called P-frames or B-frames. This encoded group of video frames is called a Group of Pictures (GOP).

[0052] P-frames contain motion vectors of pixel blocks that have changed relative to the previous frame, as well as residual information needed to recover details not captured by the motion vectors. Motion vectors and residuals are calculated unidirectionally with reference to the previous P-frame. B-frames can be considered special P-frames. Similar to P-frames, they calculate motion vectors and residuals bidirectionally with reference to the previous and next P-frames. However, since B-frames contain information from future moments, they are not very helpful for the gait recognition task itself. Therefore, this invention only uses P-frames for feature extraction.

[0053] The P-frame consists only of the motion vector V(t) at this moment and the residual Δ. i The P-frame in the video frame is composed of (t), and the t-th P-frame is reconstructed from the (t-1)-th P-frame or I-frame. The image definition of the P-frame is shown in formula (1):

[0054]

[0055] in, This represents the z-th pixel in frame t.

[0056] Motion estimation (motion vectors) and motion compensation (residuals) can remove inter-frame redundancy, greatly improving the training efficiency of deep learning methods based on video-encoded data. However, since P-frames only retain the difference from the previous frame (I-frame or P-frame), when the redundancy between adjacent frames is too high, P-frames contain almost no useful information, resulting in highly sparse data that is not conducive to network learning. Residual data and motion vectors from encoded videos are extracted from P-frames, clearly representing the human body's contours and gait features while being highly correlated with motion vector information. Directly using motion vectors extracted from gait videos often contains a lot of noise and interference, which ultimately affects recognition performance.

[0057] To address the above issues, this application proposes a gait recognition feature extraction method based on video coded data, which combines motion vectors and residual data to compensate for the deficiencies in the coded data and optimizes the coded data. The method is described below through specific embodiments.

[0058] Example 1

[0059] Figure 1 This is a schematic diagram of a gait recognition feature extraction method based on video encoded data in one embodiment of this application, as shown below. Figure 1 As shown, the gait recognition feature extraction method based on video encoded data in this embodiment can be applied to an image recognition system, including:

[0060] S1. Acquire video image data of the target to be detected, wherein the video image data is encoded using the MPEG-4 video compression standard;

[0061] S2. For the P-frame images of the image group in the video image data, based on the preset residual image entropy threshold, select residual images whose residual image entropy is greater than the residual image entropy threshold as residual images to be optimized, and use the corresponding motion vectors as motion vectors to be optimized.

[0062] S3. Based on a preset segmentation threshold, a gait contour map is obtained by segmenting from the residual image to be optimized;

[0063] S4. Optimize the motion vector to be optimized based on the gait profile map to obtain the optimized motion vector.

[0064] The gait recognition feature extraction method based on video encoded data in this embodiment selects more representative encoded data for model training and recognition by calculating the image entropy of each residual image, thereby reducing the amount of data while improving recognition accuracy and solving the problem of data sparsity. It also uses the relatively fine contour information in the residual to remove noise information caused by incorrect estimation in the motion vector, resulting in a more refined feature representation that solves the noise brought by the motion vector. This reduces the amount of computation and the utilization of computing resources in the gait recognition process, while improving the speed and accuracy of gait recognition and reducing storage overhead.

[0065] To better understand the present invention, the steps in this embodiment will be described in detail below.

[0066] S1. Acquire video image data of the target to be detected. The video image data is encoded using the MPEG-4 video compression standard.

[0067] The video image data here can be video data that is acquired and recorded in real time by a data acquisition device located in the environment of the target to be detected, or it can be pre-recorded video files stored in a video storage system.

[0068] This embodiment does not specifically limit the source of video image data.

[0069] S2. For the P-frame images of the image group in the video image data, based on the preset residual image entropy threshold, select residual images whose residual image entropy is greater than the residual image entropy threshold as residual images to be optimized, and use the corresponding motion vectors as motion vectors to be optimized.

[0070] Image entropy is a statistical form of a feature that can be used to calculate the amount of information contained in the residual. By calculating the image entropy value H of each residual image and comparing it with a pre-set threshold, the more critical P-frames can be extracted.

[0071] The specific process is as follows:

[0072] Each GOP contains 7 frames, and the residual sequence in the i-th GOP is shown in Equation (2):

[0073]

[0074] in, Let be the j-th residual map in the i-th GOP, and k represent the total number of frames in the video.

[0075] Calculate the image entropy of the residual data from the residual data;

[0076] To speed up the calculation, a weighted average method as shown in formula (3) is used to obtain the three-channel residual plot. Convert to grayscale, and then scale the image to 100x100 pixels.

[0077] Gray(x,y)=0.299Red(x,y)+0.587Green(x,y)+0.114Blue(x,y)(3)

[0078] Where Gray is a grayscale image, Red, Green, and Blue are the residual images of the red, green, and blue channels, respectively, and x and y are the horizontal and vertical coordinates of the pixel.

[0079] Then, the image entropy of each residual is calculated according to formula (4):

[0080]

[0081] in, Let be the image entropy value of the k-th residual image in the i-th GOP, and M and N be the horizontal and vertical lengths of the residual image, respectively. Calculated according to formula (5):

[0082]

[0083] Where f is the image grayscale function, Let (m, n) be the pixel value of point (m, n) in the k-th residual map of the i-th GOP.

[0084] The residual image can be filtered based on the calculated image entropy value. In this embodiment, the residual image entropy threshold is set to 1, meaning that the residual image is retained. The residual plot.

[0085] After the above screening, this embodiment filters out approximately 20% of the low-information prediction frames (P-frames) while fully preserving the information of the starting I-frame of each GOP group. The remaining image frames are then processed using the corresponding encoding compression algorithm to extract the difference information between them and the corresponding frames in the current GOP group. The filtered encoded data (motion vectors and residuals) is accumulated in GOP units. By selecting the prediction frames, the computational cost does not exceed 5% of the total model computation, but it reduces the computation of irrelevant gait by approximately 20%, allowing the network to focus on learning key features and thus improving the effective utilization of features.

[0086] S3. Based on a preset segmentation threshold, a gait contour map is obtained by segmenting from the residual image to be optimized.

[0087] To reduce the impact of noise in the residuals on gait segmentation, bilateral filtering as shown in formula (6) is used to preprocess the residual image to be optimized. Since the three-channel data format is complex and the calculation speed is slow, this embodiment performs filtering processing on the residual image after grayscale conversion according to formula (3).

[0088]

[0089] Where η is the normalization factor, Let be the filtered image, Δ(p) be the image before filtering, q be the nearest points of pixel p, Ω represent the spatial range, and w(x) and σ(x) are both Gaussian functions, with w(x) being the spatial kernel and σ(x) being the range kernel.

[0090] The normalization factor η is calculated using formula (7).

[0091] η=∫ Ω w(q)σ(Δ(pq)-Δ(p))dq (7)

[0092] The kernel function of bilateral filtering is a combination of the spatial domain and the pixel range domain. On the one hand, in flat areas of the image, pixel values ​​change little, and the corresponding pixel range domain weights are close to 1. In this case, the spatial domain weights play a major role, equivalent to Gaussian blurring. On the other hand, in edge regions of the image, pixel values ​​change significantly, and the corresponding pixel range domain weights increase, thus preserving edge information. Therefore, bilateral filtering can preserve fine gait edges and achieve noise reduction and smoothing effects.

[0093] After preprocessing the residual data, it is necessary to segment the human image from the residual. The segmentation function is shown in (8):

[0094]

[0095] Among them, B k (x,y) represents the gait profile segmented from the residual, Δ k (x,y) represents the pixel value at coordinates (x,y) in the residual map, and T represents the segmentation threshold.

[0096] Preferably, the segmentation threshold ranges from (3, 8).

[0097] S4. Optimize the motion vector to be optimized based on the gait profile map to obtain the optimized motion vector.

[0098] In step S4, considering the information features carried by the motion vector and the residual, namely: on the one hand, the motion vector contains noise and inaccurate motion, which reduces the performance of gait recognition; on the other hand, the residual has relatively fine gait contour information and is strongly correlated with the motion vector; the residual is used to obtain a fine contour map to remove noise information other than the human image in the motion vector, so that a finer motion vector can be obtained from the compression domain, thereby improving the recognition effect of the motion vector.

[0099] Specifically, after obtaining the accurate gait contour map, i.e., the residual image of the human figure, it is aligned with the motion vector data to be optimized, expanded into a dual-channel model, and the motion vector and residual are calculated according to formula (9) as follows:

[0100]

[0101] Where λ represents the optimization weight, mv k (x,y) represents the motion vector to be optimized. This represents the optimized motion vector.

[0102] By performing a dot product with the corresponding residual gait profile and adding it to the original motion vector, the motion information within the gait in the motion vector can be enhanced, thus obtaining a more effective feature representation.

[0103] This embodiment addresses the sparsity of motion vectors and residuals. First, it performs Key Coding Frame Selection (KCFS) on the residual image by using a residual image entropy threshold, selecting coding frames with higher information content to solve the sparsity problem. Second, it proposes a Joint Residual Feature Representation (JRFR) method, using the residuals to obtain a fine contour map to remove noise information other than the human figure in the motion vector, thus obtaining the object's movement speed and direction. Finally, it obtains a more refined motion vector from the compressed domain, improving the recognition effect of motion vectors.

[0104] To verify the impact of the method of this invention on gait feature storage, 10, 20, and 40 frames were taken from the CASIA-B dataset, and their storage requirements were statistically analyzed for comparison. Compared with RGB images, the method based on encoded video (MPEG-4) reduced storage by an average of 86.7%. By using the residual image entropy threshold to select key encoded frames from the residual image, the storage requirements were reduced by an average of 15% compared to the encoded video method, and by an average of 89% compared to the original RGB method.

[0105] When measuring the optimization time of different methods for a single frame / video using a single thread on an i5-9600K 3.70GHz CPU, it was found that KCFS only requires 4.6ms to calculate the entropy value of a single frame, which can improve the sparsity problem of motion vectors and residuals, while JRFR requires 8.7ms for single-frame MV optimization. To ensure fairness in testing videos with different frame numbers, the original dataset was split into multiple 10-frame videos for testing. The single-video optimization time of KCFS+JRF is less than ten times that of single-frame processing.

[0106] The following provides a specific example in conjunction with the method of this embodiment.

[0107] In this example, only video sequences with the same 90° viewpoint from the CASIA-B dataset are used. The first four normal walking gait sequences (nm-01, nm-02, nm-03, nm-04) are used as training data, and the remaining sequences are used for testing. The specific experimental design of the dataset is shown in Table 2. Table 2 is the dataset table used in this example.

[0108] Table 2

[0109]

[0110] First, all video data in CASIA-B is encoded using MPEG-4, and residuals are extracted from the encoded videos. The residuals are then read sequentially within the dataset, and the image entropy of each residual is calculated according to formula (4).

[0111] It should be noted that, in order to verify the technical effect of the method in this application, the existing dataset was encoded with MPEG-4. In actual use, the MPEG-4 bitstream can be directly extracted from the encoder that encodes the real-time video stream at the front end.

[0112] The residual map is filtered based on the calculated image entropy value, discarding frames with excessive redundant information while retaining frames with more information. The specific filtering algorithm is shown in Table 1, which is a table of key coded frame selection algorithms, where `threshold` is the filtering threshold.

[0113] Table 1

[0114]

[0115] Then, joint residual feature optimization is performed, where the segmentation threshold is set to 6, and a more refined gait profile is extracted from the residual to optimize the motion vector. Figure 2 This is a network model framework diagram for gait recognition in one embodiment of this application, such as... Figure 2 As shown, this network model framework uses ResNet-50 as the backbone to model I-frames and uses the smaller and simpler ResNet-18 as the backbone to capture changes in P-frames.

[0116] In the I-frame model, the input video encoding is divided into several GOP sequences, and then a frame is randomly selected from each GOP to fully utilize the information contained in the video encoding. On the ResNet-50 network backbone, to promote the interaction between long-term and short-term motion information, this embodiment inserts a time-transfer module into each standard residual block. Furthermore, to capture the interaction between local cross-channels, this paper also inserts an Efficient Channel Attention (ECA) module to improve the network model's performance.

[0117] The KPFS and JRFR modules mentioned earlier were inserted into the P-frame model to optimize the extracted feature data. Finally, the ECA module was retained on the ResNet-18 network backbone. Furthermore, to enhance the shallow network's ability to extract long-term temporal features, a Gate Recurrent Unit (GRU) module was added before the fully connected layers, resulting in better performance on long sequences.

[0118] By leveraging the relatively fine gait contour features in the corresponding residuals, noise information other than gait is removed from the motion vector, significantly improving the recognition accuracy of the motion vector.

[0119] During the training phase, the epochs were set to 100 and the learning rate was 0.0001. All keyframes in nm(01-04) were fed into the network model for training.

[0120] During the testing phase, nm(05-06), bg(01-02), and cl(01-02) were used for testing, and the average recognition accuracy was 93.8%.

[0121] Example 2

[0122] A second aspect of this application provides a gait recognition method. Figure 3 This is a schematic diagram of the gait recognition method in another embodiment of this application, as shown below. Figure 3 As shown, the method includes:

[0123] The optimized motion vector of the target to be detected is obtained by the gait recognition feature extraction method based on video encoded data as described in any of the above embodiments;

[0124] The optimized motion vector is input into a pre-established gait recognition model to obtain the identity recognition information of the target to be detected.

[0125] The gait recognition method provided in this embodiment can be used to perform the steps of the gait recognition feature extraction method based on video encoded data in the above method embodiment. Its implementation principle and technical effect are similar, and will not be described again in this embodiment.

[0126] Example 3

[0127] A third aspect of this application provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor. When executed by the processor, the computer program implements the steps of the gait recognition feature extraction method based on video encoded data as described in any of the above embodiments.

[0128] Figure 4 This is a schematic diagram of the architecture of an electronic device in another embodiment of this application.

[0129] Figure 4 The illustrated electronic device may include at least one processor 101, at least one memory 102, at least one network interface 104, and other user interfaces 103. The various components of the electronic device are coupled together via a bus system 105. It is understood that the bus system 105 is used to implement communication between these components. In addition to a data bus, the bus system 105 also includes a power bus, a control bus, and a status signal bus. However, for clarity, in… Figure 4 The general labeled all buses as Bus System 105.

[0130] The user interface 103 may include a display, keyboard, or clicking device (e.g., mouse, trackball, or touchpad).

[0131] It is understood that the memory 102 in this embodiment can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDRSDRAM), Enhanced Synchronous DRAM (ESDRAM), Sync Link DRAM (SLDRAM), and Direct Rambus RAM (DRRAM). The memory 62 described herein is intended to include, but is not limited to, these and any other suitable types of memory.

[0132] In some implementations, memory 102 stores elements, executable units or data structures, or subsets thereof, or extended sets thereof: operating system 1021 and application program 1022.

[0133] The operating system 1021 includes various system programs, such as a framework layer, a core library layer, and a driver layer, used to implement various basic business functions and handle hardware-based tasks. The application program 1022 includes various applications used to implement various application functions. Programs implementing the methods of this embodiment of the invention can be included in the application program 1022.

[0134] In this embodiment of the invention, the processor 101 executes the method steps provided in the first aspect by calling the program or instructions stored in the memory 102, specifically the program or instructions stored in the application program 1022.

[0135] The methods disclosed in the above embodiments of the present invention can be applied to processor 101, or implemented by processor 101. Processor 101 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the integrated logic circuit of the hardware in processor 101 or by instructions in the form of software. The processor 101 may be a general-purpose processor, digital signal processor, application-specific integrated circuit, off-the-shelf programmable gate array or other programmable logic device, discrete gate or transistor logic device, or discrete hardware component. It can implement or execute the methods, steps and logic block diagrams disclosed in the embodiments of the present invention. The general-purpose processor may be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of the present invention can be directly embodied as being executed by a hardware decoding processor, or executed by a combination of hardware and software units in the decoding processor. The software units may be located in random access memory, flash memory, read-only memory, programmable read-only memory or electrically erasable programmable memory, registers and other mature storage media in the art. The storage medium is located in memory 102, and processor 101 reads the information in memory 102 and completes the steps of the above method in combination with its hardware.

[0136] In addition, in conjunction with the gait recognition feature extraction method based on video encoded data in the above embodiments, the present invention can provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements any of the gait recognition feature extraction methods based on video encoded data in the above method embodiments.

[0137] It should be noted that any reference numerals placed between parentheses in the claims should not be construed as limiting the claims. The word "comprising" does not exclude the presence of components or steps not listed in the claims. The words "a" or "an" preceding a component do not exclude the presence of a plurality of such components. The invention can be implemented by means of hardware comprising several different components and by means of a suitably programmed computer.

[0138] Furthermore, it should be noted that in the description of this specification, the terms "one embodiment," "some embodiments," "embodiment," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Furthermore, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0139] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the claims should be interpreted to include both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0140] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, then this invention should also include these modifications and variations.

Claims

1. A gait recognition feature extraction method based on video encoding data, characterized in that, The method includes the following steps: S1. Acquire video image data of the target to be detected, wherein the video image data is encoded using the MPEG-4 video compression standard; S2. For the P-frame images of the image group in the video image data, based on the preset residual image entropy threshold, select residual images whose residual image entropy is greater than the residual image entropy threshold as residual images to be optimized, and use the corresponding motion vectors as motion vectors to be optimized. S3. Based on a preset segmentation threshold, a gait contour map is obtained by segmenting from the residual image to be optimized; S4. Optimize the motion vector to be optimized based on the gait profile map to obtain the optimized motion vector.

2. The gait recognition feature extraction method based on video encoded data according to claim 1, characterized in that, Step S2 includes: The residual images of the three channels in the P-frame image are converted into grayscale images using a weighted average method. Calculate the image entropy of each grayscale residual image; Grayscale residual images with residual image entropy greater than a preset residual image entropy threshold are selected as residual images to be optimized, and the corresponding motion vectors are selected as motion vectors to be optimized.

3. The gait recognition feature extraction method based on video encoded data according to claim 2, characterized in that, The residual image of the three channels in the P-frame image is converted into a grayscale image using the following formula, and then converted into... size: ； Wherein, Gray is a grayscale image, Red, Green, and Blue are the residual images of the red, green, and blue channels, respectively, x and y are the horizontal and vertical coordinates of the pixel, Gray(x,y) is the pixel value of the grayscale image at coordinates (x,y), Red(x,y) is the pixel value of the red channel residual image at coordinates (x,y), Green(x,y) is the pixel value of the green channel residual image at coordinates (x,y), and Blue(x,y) is the pixel value of the blue channel residual image at coordinates (x,y).

4. The gait recognition feature extraction method based on video encoded data according to claim 1, characterized in that, The steps following step S2 and before step S3 also include: The residual image to be optimized is subjected to bilateral filtering using the following formula. ； in, As the normalization factor, This is the filtered image. The image is before filtering, where q represents the nearest points of pixel p, and Ω represents the spatial range. and Both are Gaussian functions. For space core, For the range kernel.

5. The gait recognition feature extraction method based on video encoded data according to claim 1, characterized in that, Step S3 includes: based on a preset segmentation threshold, using the following segmentation function to segment the residual image to be optimized to obtain a gait contour map: ； in, This represents the gait profile segmented from the residual. Represents the coordinates corresponding to the residual plot pixel values, T This represents the segmentation threshold.

6. The gait recognition feature extraction method based on video encoded data according to claim 1, characterized in that, Step S4 includes: The gait profile is aligned with the motion vector to be optimized and expanded into a dual-channel model. Based on the gait profile, the motion vector to be optimized is optimized according to the following formula to obtain the optimized motion vector: ； in, This indicates the optimization weights. This represents the motion vector to be optimized. This represents the optimized motion vector. This represents the gait profile segmented from the residual.

7. The gait recognition feature extraction method based on video encoded data according to claim 1, characterized in that, The preset residual image entropy threshold is set to 1.

8. A gait recognition method, characterized in that, The method includes: The optimized motion vector of the target to be detected is obtained by the gait recognition feature extraction method based on video encoded data as described in any one of claims 1 to 7 above. The optimized motion vector is input into a pre-established gait recognition model to obtain the identity recognition information of the target to be detected.

9. An electronic device, characterized in that, include: The memory, the processor, and the computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of the gait recognition feature extraction method based on video coded data as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the gait recognition feature extraction method based on video encoded data as described in any one of claims 1 to 7.

Images