A gait recognition method based on the fusion of human body contour and key point features

Abstract

The present invention discloses a gait recognition method based on the fusion of human body contour and key point features, comprising: inputting a gait video of a single person walking, obtaining a pedestrian contour sequence in the video; substituting the gait video into an OpenPose algorithm module, obtaining a normalized human body key point information sequence, substituting the pedestrian contour sequence into a GaitSet algorithm module, obtaining the features of the gait contour sequence; extracting the human body key point information sequence into a human body key point feature extraction module composed of LSTM and CNN; obtaining a gait contour feature vector and a human body key point feature vector respectively; connecting the gait contour feature vector and the human body key point feature vector and inputting them into a feature fusion module; importing the gait fusion feature into a fusion network for feature learning, and identifying the identity of the person in the video. The present invention utilizes a human body contour feature extraction module and a human body key point feature extraction module to extract their features respectively, and then performs feature layer fusion to obtain a gait fusion feature, thereby improving the accuracy and robustness of gait recognition.

Description

Technical Field

[0001] The present invention belongs to the cross-technical fields of computer vision, identity recognition, feature fusion, etc., and specifically relates to a gait recognition method based on the fusion of human body contour and key point features. Background Art

[0002] Gait is a kind of human biological behavior characteristic, which describes the movement and change rules of the upper and lower joints during walking. Medically, it is believed that each person's gait is unique. The human gait characteristics have the characteristics of global uniqueness, long-term invariance, non-reversibility, easy collection, non-imitability, non-disguiseability, and non-contact. It is currently the most suitable biometric feature for widespread popularization in multiple fields. Gait recognition is to extract physical and behavioral characteristics from the individual's walking pattern for identity recognition. In the field of advanced security, due to the difficulty of camouflage and imitation of gait characteristics, iris characteristics can be used to assist in hybrid biometric recognition to improve security performance, bringing stronger security protection to financial and military fields.

[0003] Although each person's gait is unique, factors such as clothing, things carried, and viewing angles all pose major challenges to gait recognition. In order to solve these problems, people have proposed various methods. In the early days, people distinguished people from the differences in various parts of the body when each person walked, but this meant modeling many structures of the body, involving a large number of variables and complex calculations; recognition from raw RGB images is also a research direction, but this method faces the challenge of eliminating irrelevant gait information; with the rapid development of deep learning, the two gait recognition methods based on human contours and human skeleton key points have gradually become mainstream.

[0004] The method based on human body contour can largely avoid the interference of irrelevant pixels in the video sequence, quickly and effectively extract features from the contour sequence, and this method is suitable for low-resolution conditions. Although the method based on contour sequence has many advantages, this method only retains the external contour of the human body, and the torso information of the person cannot be used during walking. The method based on human skeleton key points estimates the posture of the person in the video sequence, extracts the skeleton key points of the person, and identifies them from the skeleton key point sequence. This method can extract body structure information well, but ignores the identifiable body shape information, resulting in low performance.

[0005] The human silhouette retains body shape information, and the human skeleton key points retain body structure information. These two features are complementary, and their composition is expected to become a more comprehensive representation of gait. However, the complementary advantages of the silhouette and skeleton key points have not been fully utilized in previous studies. Summary of the invention

[0006] Purpose of the invention: In order to overcome the deficiencies in the prior art, the present invention provides a gait recognition method based on the fusion of human body contour and key point features. The method extracts a human body contour sequence and a human body key point sequence from a pedestrian gait video, respectively, and uses a human body contour feature extraction module and a human body key point feature extraction module to extract their features respectively, and then performs feature layer fusion to obtain gait fusion features, thereby improving the accuracy and robustness of gait recognition.

[0007] Technical solution: In the first aspect, the present invention provides a gait recognition method based on the fusion of human body contour and key point features, comprising:

[0008] Based on the input single-person walking gait video, obtain the pedestrian outline sequence in the video, and substitute the gait video into the OpenPose algorithm module to obtain the normalized human key point information sequence;

[0009] Substitute the pedestrian contour sequence into the GaitSet algorithm module to obtain the features of the gait contour sequence; import the human key point information sequence into the human key point feature extraction module to obtain the features of the human key points;

[0010] Based on the features of the gait contour sequence and the features of the key points of the human body, a gait contour feature vector and a human key point feature vector are obtained respectively;

[0011] The gait contour feature vector and the human body key point feature vector are connected and input into the feature fusion module to obtain the gait fusion feature;

[0012] The gait fusion features are imported into the fusion network for feature learning to identify the identity of the person in the video.

[0013] In a further embodiment, a method of inputting a gait video of a single person walking and obtaining a pedestrian profile sequence in the video includes:

[0014] The KNN algorithm is used in the gait video to obtain the human body contour of each frame of the gait video;

[0015] Based on the human body contour of each frame of the image, the number of non-zero pixels of each single person contour image is calculated, and whether to output the image is selected according to the threshold value of the image pixels;

[0016] Obtaining the index value interval of the upper limit value and the lower limit value of the row pixel sum that is not 0 for the output image, and cropping the upper and lower areas of the output image according to the index value interval to obtain a cropped image;

[0017] Find the median based on the x-axis in the cropped image, and determine the x-axis center point of the person in the image by finding the median;

[0018] Slice from the center point to both sides to obtain an image array with 64-bit rows and columns;

[0019] Convert the image array type to obtain a sequence of pedestrian silhouettes.

[0020] In a further embodiment, the method of substituting the gait video into the OpenPose algorithm module to obtain a normalized human key point information sequence includes:

[0021] Based on the gait video, obtain the position coordinates of each key point of the human body in the video;

[0022] The position of the neck key point is selected as the origin in the position coordinates of each key point of the human body in the video, and the distance between the neck and the hip is used as a reference to normalize other key points to obtain a normalized human key point frame sequence;

[0023] Among them, the normalization formula is:

[0024]

[0025] Where P i is the position of key point i, P′ i is the normalized position of the key point, P is the position of the neck joint point, and D is the distance between the neck and hip key points.

[0026] In a further embodiment, the human body key point feature extraction module includes: an LSTM module and a CNN module; the human body key point information sequence is imported into the human body key point feature extraction module, and the method for obtaining the characteristics of the human body key points is to respectively pass the acquired human body key point frame sequence into the LSTM module and the CNN module to obtain the characteristics of each frame of the human body key points.

[0027] In a further embodiment, based on the features of the human body key points, a method for obtaining a human body key point feature vector includes:

[0028] Based on the features of each frame of the key points of the human body, a feature vector associated with the features of each frame is obtained, and the obtained feature vectors of each frame are connected;

[0029] The concatenated feature vector is input into the compression block to obtain a 62╳128-dimensional feature vector of human key points.

[0030] In a further embodiment, the LSTM model consists of a fully connected layer and an LSTM layer, and the feature dimension of the LSTM is set to 256 dimensions;

[0031] The CNN module is configured with 10 3×3 convolutional layers, the number of filters of the first convolutional layer is set to 32, a pooling layer is set between the second and third convolutional layers and the fifth and sixth convolutional layers, the number of filters of the second to fourth convolutional layers is set to 64, the number of remaining filters is set to 128, and the first pooling layer is residually connected to the fourth convolutional layer, the second pooling layer is residually connected to the seventh convolutional layer, and the dimension of the fully connected layer is set to 256 dimensions;

[0032] The feature extraction module also includes a compression module, which consists of a BN layer, a ReLU layer, a Dropout layer and a 128-dimensional fully connected layer.

[0033] In a further embodiment, the gait profile feature vector and the human body key point feature vector are connected and input into a feature fusion module, and the method for obtaining the gait fusion feature includes:

[0034] Connect each dimension of the gait profile feature vector and the human body key point feature vector to obtain a connection vector of each dimension;

[0035] Import the connection vector of each dimension into the fully connected layer of the feature fusion module to obtain the human gait fusion feature vector;

[0036] Among them, the feature fusion module introduces a ternary loss function for training. The expression formula of the ternary loss function is:

[0037]

[0038]

[0039] Where, L BA () represents the sum of the loss values ​​of positive and negative samples, Represents the difference between the distance between the anchor sample and the positive sample and the distance between the anchor sample and the negative sample. represents the i-th anchor point sample, represents the jth anchor point sample, D represents the distance between two samples, represents the i-th positive sample, represents the i-th negative sample, m represents the threshold parameter set according to actual needs, which is used to control the difference between the distance between the anchor sample and the positive sample and the distance between the anchor sample and the negative sample, a represents the anchor sample, p represents the negative sample, n represents the negative sample, i represents the positive sample number, j represents the negative sample number, P represents P IDs, and K represents that each ID has K samples.

[0040] In a further embodiment, the method of importing the gait fusion feature into the fusion network for feature learning and identifying the identity of the person in the video includes:

[0041] The gait fusion feature F Q And each feature F in the fusion network feature library G Calculate the Euclidean distance to obtain the gait fusion feature F Q And each feature F in the fusion network feature library G Distance results;

[0042] Based on the distance results, an approximate distance result is selected and a recognition result is determined based on features associated with the approximate distance results, thereby completing the identity recognition of the person in the video.

[0043] In a second aspect, the present invention provides a processing device, including a memory and a processor, wherein the memory stores a computer program, which is executed by the processor to implement the above-mentioned gait recognition method based on the fusion of human body contour and key point features.

[0044] In a third aspect, the present invention provides a readable storage medium having a computer program stored thereon, which implements the steps of the above method when executed by a processor.

[0045] Beneficial effects: Compared with the prior art, the present invention has the following advantages:

[0046] (1) The present invention uses morphological methods and OpenPose algorithm to extract the contour sequence and skeleton key point sequence of the characters in the gait video respectively as the initial walking feature representation of the pedestrians.

[0047] (2) The present invention introduces the GaitSet algorithm to extract features in the contour sequence, and then extracts the temporal features in the skeleton key point sequence through a combination of an LSTM network and a CNN network.

[0048] (3) The present invention uses a feature layer fusion method to obtain the fusion features of gait, and obtains a comprehensive feature expression of each person through learning of the fusion network, thereby effectively improving the accuracy and reliability of gait recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 is a flow chart of the overall method.

[0050] Figure 2 This is the architecture diagram of the gait feature extraction network based on feature fusion. DETAILED DESCRIPTION

[0051] In order to more fully understand the technical content of the present invention, the technical solution of the present invention is further introduced and illustrated in conjunction with specific embodiments below, but is not limited thereto.

[0052] like Figure 1 , Figure 2A gait recognition method based on the fusion of human body contour and key point features is shown, comprising the following steps:

[0053] Step 1) Input a single person walking gait video and obtain the pedestrian profile sequence in the video

[0054] In the process of obtaining the pedestrian contour sequence in the video, the pedestrian contour sequence is also cropped to remove invalid pixels. The specific steps are as follows:

[0055] Step 11) using KNN algorithm to extract the human body contour of each frame image;

[0056] Step 12) Calculate the number of non-zero pixels in each frame of the single person outline image. If the sum of the image pixels is less than 10,000, do not return the image information;

[0057] Step 13) Obtain the highest and lowest indexes of the row pixel sums that are not 0, and use them to crop the upper and lower areas of the image;

[0058] Step 14) Get the median of the x-axis and regard it as the x-center of the person. If the median of the image cannot be found, no image information is returned;

[0059] Step 15) Slice from the center point to both sides to obtain an image array with 64 rows and columns. If the value exceeds the image range, shift it to the right and then splice it from both sides through an all-0 array;

[0060] Step 16) Convert the image array type and return it to obtain a cropped pedestrian silhouette sequence.

[0061] Step 2: Substitute the gait video into the OpenPose algorithm module to obtain a normalized human key point information sequence, and substitute the pedestrian contour sequence into the GaitSet algorithm module to obtain the features of the gait contour sequence;

[0062] The steps to obtain the normalized human key point information sequence are as follows:

[0063] Step 21) obtaining the position coordinates of each key point of the human body in the video;

[0064] Step 22) In the key point detection of the human body, the neck and hips are relatively stable key points, and the position of the neck joint point is selected as the origin;

[0065] Step 23) Based on the distance between the neck and the hip, normalize the other key points. The normalization formula is:

[0066]

[0067] Where P i is the position of key point i, P′ iis the normalized position of the key point, P is the position of the neck, and D is the distance between the neck and hip key points.

[0068] Secondly, the pedestrian contour feature extraction includes the following steps:

[0069] Step 24) inputting the pedestrian contour sequence obtained in step 1) into the GaitSet network to obtain pedestrian contour features;

[0070] Step 25) After calculation by the GaitSet network, a 62*128-dimensional feature vector is obtained.

[0071] Step 3: The human body key point information sequence is extracted by a human body key point feature extraction module composed of LSTM and CNN to obtain the features of the human body key points;

[0072] Step 4: Based on the features of the key points of the human body, the method for obtaining the feature vector of the key points of the human body includes:

[0073] Step 41) The human body key point frame sequence obtained is respectively passed into the LSTM module and the CNN module to obtain the features of each frame of the human body key point, and the feature vectors of each frame obtained are connected;

[0074] Step 42) The concatenated feature vector is input into the compression block to obtain a 62×128-dimensional feature vector of the key points of the human body.

[0075] The human body key point feature extraction module consists of an LSTM network for extracting temporal information, a CNN network for extracting spatial information, and a compression module.

[0076] The LSTM model consists of a fully connected layer and an LSTM layer, and the feature dimension of the LSTM is set to 256 dimensions.

[0077] The CNN model has 10 3×3 convolutional layers. The number of filters in the first convolutional layer is set to 32. A pooling layer is set between the second and third convolutional layers and between the fifth and sixth convolutional layers. The number of filters in the second to fourth convolutional layers is set to 64, and the number of remaining filters is set to 128. Drawing on the residual connection idea of ​​ResNet, the first pooling layer is residually connected to the fourth convolutional layer, the second pooling layer is residually connected to the seventh convolutional layer, and the dimension of the fully connected layer is set to 256.

[0078] To prevent overfitting, a compression module is set up, which consists of a BN layer, a ReLU layer, a Dropout layer, and a 128-dimensional fully connected layer.

[0079] Step 5: Connect the gait contour feature vector and the human body key point feature vector and input them into the feature fusion module. The method for obtaining the gait fusion feature includes:

[0080] Step 51) connect each dimension of the 62╳128-dimensional gait profile feature vector and the 62╳128-dimensional human key point feature vector to obtain a connection vector of each dimension;

[0081] Step 52) importing the connection vector of each dimension into the fully connected layer of the feature fusion module to obtain the human gait fusion feature vector;

[0082] Among them, the feature fusion module introduces a ternary loss function for training. The expression formula of the ternary loss function is:

[0083]

[0084]

[0085] Where, L BA () represents the sum of the loss values ​​of positive and negative samples, Represents the difference between the distance between the anchor sample and the positive sample and the distance between the anchor sample and the negative sample. represents the i-th anchor point sample, represents the jth anchor point sample, D represents the distance between two samples, represents the i-th positive sample, represents the i-th negative sample, m represents the threshold parameter set according to actual needs, which is used to control the difference between the distance between the anchor sample and the positive sample and the distance between the anchor sample and the negative sample, a represents the anchor sample, p represents the negative sample, n represents the negative sample, i represents the positive sample number, j represents the negative sample number, P represents P IDs, and K represents that each ID has K samples.

[0086] Step 6: Import the gait fusion features into the fusion network for feature learning. The method for identifying the identity of the person in the video includes:

[0087] Step 61) Fusion gait feature F Q And each feature F in the fusion network feature library G Calculate the Euclidean distance to obtain the gait fusion feature F Q And each feature F in the fusion network feature library G Distance results;

[0088] Step 62) Based on the distance results, an approximate distance result is selected and a recognition result is determined based on features associated with the approximate distance results, thereby completing identity recognition of the person in the video.

[0089] The present invention obtains a gait video sequence, extracts a human body contour sequence and a human body key point sequence from the video sequence respectively, extracts their features respectively using a human body contour feature extraction module and a human body key point feature extraction module, then performs feature layer fusion to obtain gait fusion features, performs feature learning through a fusion network, and introduces a ternary loss function for training, thereby realizing accurate identification of the identity of the person in the video.

[0090] Embodiment 2 provides a processing device, including a memory and a processor, wherein the memory stores a computer program, which is executed by the processor to implement the following gait recognition method based on the fusion of human body contour and key point features:

[0091] Based on the input single-person walking gait video, obtain the pedestrian outline sequence in the video, and substitute the gait video into the OpenPose algorithm module to obtain the normalized human key point information sequence;

[0092] Substitute the pedestrian contour sequence into the GaitSet algorithm module to obtain the features of the gait contour sequence; import the human key point information sequence into the human key point feature extraction module to obtain the features of the human key points;

[0093] Based on the features of the gait contour sequence and the features of the key points of the human body, a gait contour feature vector and a human key point feature vector are obtained respectively;

[0094] The gait contour feature vector and the human body key point feature vector are connected and input into the feature fusion module to obtain the gait fusion feature;

[0095] The gait fusion features are imported into the fusion network for feature learning to identify the identity of the person in the video.

[0096] Embodiment 3 provides a readable storage medium on which a computer program is stored, and when the program is executed by a processor, the steps of the above method are implemented.

[0097] In summary, the present invention uses the morphological method and the OpenPose algorithm to respectively extract the contour sequence and the skeleton key point sequence of the character in the gait video as the initial walking feature representation of the pedestrian; introduces the GaitSet algorithm to extract the features in the contour sequence, and then extracts the temporal features in the skeleton key point sequence through the combination of the LSTM network and the CNN network; uses the feature layer fusion method to obtain the fusion features of the gait, and obtains the comprehensive feature expression of each person through the learning of the fusion network, which effectively improves the accuracy and reliability of gait recognition.

[0098] The embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of complete hardware embodiments, complete software embodiments, or embodiments in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0099] The embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of complete hardware embodiments, complete software embodiments, or embodiments in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0100] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0101] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0102] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0103] The above is only a preferred embodiment of the present invention. Without departing from the technical principle of the present invention, several improvements and modifications can be made, and these improvements and modifications should also be regarded as the protection scope of the present invention.

Claims

1. A gait recognition method based on the fusion of human body contour and key point features, characterized in that: include: Input a single person walking gait video, obtain the pedestrian contour sequence in the video, and crop the pedestrian contour sequence to remove invalid pixels; Substitute the gait video into the OpenPose algorithm module to obtain the normalized human key point information sequence, and substitute the pedestrian contour sequence into the GaitSet algorithm module to obtain the features of the gait contour sequence; The human body key point information sequence is extracted by a human body key point feature extraction module composed of LSTM and CNN to obtain the features of the human body key points; Based on the features of the gait contour sequence and the features of the key points of the human body, a gait contour feature vector and a human key point feature vector are obtained respectively; The gait contour feature vector and the human body key point feature vector are connected and input into the feature fusion module to obtain the gait fusion feature; Import the gait fusion features into the fusion network for feature learning and identify the identity of the person in the video; The gait contour feature vector and the human body key point feature vector are connected and input into the feature fusion module. The method of obtaining the gait fusion feature includes: Connect each dimension of the gait profile feature vector and the human body key point feature vector to obtain a connection vector of each dimension; Import the connection vector of each dimension into the fully connected layer of the feature fusion module to obtain the human gait fusion feature vector; Among them, the feature fusion module introduces a ternary loss function for training. The expression formula of the ternary loss function is: Where, L BA () represents the sum of the loss values ​​of positive and negative samples, Represents the difference between the distance between the anchor sample and the positive sample and the distance between the anchor sample and the negative sample. represents the i-th anchor point sample, represents the jth anchor point sample, D represents the distance between two samples, represents the i-th positive sample, represents the i-th negative sample, m represents the threshold parameter set according to actual needs, which is used to control the difference between the distance between the anchor sample and the positive sample and the distance between the anchor sample and the negative sample, a represents the anchor sample, p represents the negative sample, n represents the negative sample, i represents the positive sample number, j represents the negative sample number, P represents P IDs, and K represents that each ID has K samples.

2. The gait recognition method based on the fusion of human body contour and key point features according to claim 1 is characterized in that: Input a single person walking gait video, obtain a pedestrian contour sequence in the video, and crop the pedestrian contour sequence to remove invalid pixels. The method includes: The KNN algorithm is used in the gait video to obtain the human body contour of each frame of the gait video; Based on the human body contour of each frame of the image, the number of non-zero pixels of each single person contour image is calculated, and whether to output the image is selected according to the threshold value of the image pixels; Obtaining the index value interval of the upper limit value and the lower limit value of the row pixel sum that is not 0 for the output image, and cropping the upper and lower areas of the output image according to the index value interval to obtain a cropped image; Find the median based on the x-axis in the cropped image, and determine the x-axis center point of the person in the image by finding the median; Slice from the center point to both sides to obtain an image array with 64-bit rows and columns; Convert the image array type to obtain a sequence of pedestrian silhouettes.

3. The gait recognition method based on the fusion of human body contour and key point features according to claim 1 is characterized in that: Substituting the gait video into the OpenPose algorithm module to obtain a normalized human key point information sequence includes: Based on the gait video, obtain the position coordinates of each key point of the human body in the video; The position of the neck key point is selected as the origin in the position coordinates of each key point of the human body in the video, and the distance between the neck and the hip is used as a reference to normalize other key points to obtain a normalized human key point frame sequence; Among them, the normalization formula is: Where P i is the position of key point i, P i ' is the normalized position of the key point, P is the position of the neck joint point, and D is the distance between the neck and hip key points.

4. The gait recognition method based on the fusion of human body contour and key point features according to claim 1 is characterized in that: Based on the features of the key points of the human body, the method of obtaining the feature vector of the key points of the human body includes: The obtained human key point frame sequences are respectively passed into the LSTM module and the CNN module to obtain the features of each frame of the human key points, and the obtained feature vectors of each frame are connected; The input of the connected feature vector is compressed into blocks to obtain a 62╳128-dimensional human key point feature vector.

5. The gait recognition method based on the fusion of human body contour and key point features according to claim 4 is characterized in that: The LSTM module is composed of a fully connected layer and an LSTM layer, and the feature dimension of the LSTM module is set to 256 dimensions; The CNN module is configured with 10 3×3 convolutional layers, the number of filters of the first convolutional layer is set to 32, a pooling layer is set between the second and third convolutional layers and the fifth and sixth convolutional layers, the number of filters of the second to fourth convolutional layers is set to 64, the number of remaining filters is set to 128, and the first pooling layer is residually connected to the fourth convolutional layer, the second pooling layer is residually connected to the seventh convolutional layer, and the dimension of the fully connected layer is set to 256 dimensions; The feature extraction module also includes a compression module, which consists of a BN layer, a ReLU layer, a Dropout layer and a 128-dimensional fully connected layer.

6. The gait recognition method based on the fusion of human body contour and key point features according to claim 1, characterized in that: The method of importing the gait fusion feature into the fusion network for feature learning and identifying the identity of the person in the video includes: The gait fusion feature F Q And each feature F in the fusion network feature library G Calculate the Euclidean distance to obtain the gait fusion feature F Q And each feature F in the fusion network feature library G Distance results; Based on the distance results, an approximate distance result is selected and a recognition result is determined based on features associated with the approximate distance results, thereby completing the identity recognition of the person in the video.

7. A processing device, comprising a memory and a processor, characterized in that: The memory stores a computer program, which is executed by the processor to implement the gait recognition method based on the fusion of human body contour and key point features as described in any one of claims 1 to 6.

8. A readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images