CN115240269A

CN115240269A - Gait recognition method and device based on body type transformation and storage medium

Info

Publication number: CN115240269A
Application number: CN202210777669.6A
Authority: CN
Inventors: 王昕�; 潘华东; 殷俊
Original assignee: Zhejiang Dahua Technology Co Ltd
Current assignee: Zhejiang Dahua Technology Co Ltd
Priority date: 2022-06-30
Filing date: 2022-06-30
Publication date: 2022-10-25

Abstract

The application discloses gait recognition method, equipment and storage medium based on body type transformation, and the gait recognition method comprises the following steps: acquiring a gait image sequence, wherein the gait image sequence comprises a plurality of gait images, acquiring a target gait contour map and a target key point matrix corresponding to the same target respectively based on each gait image, and performing aspect ratio transformation on the target contour in each target gait contour map to obtain a target body type transformation contour map. By the method, the problem of low target gait recognition accuracy caused by clothes shielding, carrying objects and the like can be effectively solved.

Description

Gait recognition method and device based on body type transformation and storage medium

Technical Field

The invention relates to the field of biological identification, in particular to a gait identification method and device based on body type transformation and a storage medium.

Background

The physiological characteristics of the human body mainly include fingerprints, palm prints, facial characteristics and the like, and the behavior characteristics mainly include signature handwriting of gait characteristics and voice characteristics and the like. In recent years, biometric identification technology based on gait characteristics has been increasingly used. The gait recognition is to extract features and recognize gait based on the moving target extraction, and the gait recognition technology can complete the acquisition under the remote and non-contact condition to realize the identity recognition under the non-contact and remote condition. Although the research on the gait recognition technology is more and more intensive and comprehensive, the current gait recognition still has the problem of low gait recognition accuracy on gait images due to the change of visual angles (different postures of human bodies and differences of collected gait characteristics under different visual angles), the change of clothes (the outlines of the legs of the human bodies are shielded due to overlong/too thick clothes, and the like), the change of carried objects (the integrity of the gait outline map is influenced by carrying factors such as a backpack/handbag, and the like), and the like.

Disclosure of Invention

The application provides a gait recognition method, equipment and a storage medium based on body type transformation, and the problem of low gait recognition accuracy caused by factors such as clothing shielding in gait recognition can be effectively solved by acquiring a target gait contour map, a target key point matrix and a target body type transformation contour map.

In order to solve the technical problem, the present application adopts a technical solution that: provided is a gait recognition method based on body type transformation, which comprises the following steps: acquiring a gait image sequence, wherein the gait image sequence comprises a plurality of gait images; respectively acquiring a target gait contour map and a target key point matrix corresponding to the same target based on each gait image; carrying out aspect ratio conversion on target contours in each target gait contour map to obtain a target body type conversion contour map; acquiring gait space-time characteristics by using a gait space-time extraction network based on the target gait contour map and the target body type transformation contour map; based on the target key point matrix, extracting a network by using key point features to obtain target key point features; and performing feature fusion on the gait space-time features and the target key point features to obtain fusion measurement features.

Wherein, acquire gait image sequence, include: acquiring an original image sequence formed by shooting a target in a walking process, wherein the original image sequence comprises a plurality of original images; respectively carrying out target detection and target tracking on the original image; and selecting an original image with the target leg part not being blocked from the plurality of original images to form a gait image sequence.

The method for obtaining the target body type transformation contour map by carrying out aspect ratio transformation on the target gait contour map comprises the following steps: inputting the original aspect ratio of the target contour of each target gait contour map into a body type transformation function to obtain a corresponding target aspect ratio; or based on the expectation and standard deviation of the original aspect ratio set of the target profile of the target gait profile to obtain a corresponding target aspect ratio; and transforming the target contour based on the target width-height ratio.

The method for inputting the original aspect ratio of the target contour of each target gait contour map into the body type transformation function comprises the following steps: calculating one of a maximum value, a minimum value, a mode, a mean value or a median of the plurality of original aspect ratios as the target aspect ratio by using a body type transformation function; or obtaining the expectation and standard deviation of the normally distributed original aspect ratio set, and then selecting the target aspect ratio from the value range formed by the expectation and standard deviation.

Wherein, the target aspect ratio corresponding to each target gait contour map is the same or different.

Wherein, based on the target gait contour map and the target body type transformation contour map, the gait space-time extraction network is used to obtain the gait space-time characteristics, which comprises the following steps: splicing the corresponding target gait contour map and the target body type transformation contour map according to the channel dimension to form a fusion contour map; and extracting gait space-time characteristics from the fusion contour map by using a gait space-time extraction network.

Based on the target key point matrix, the method for acquiring the target key point features by using the key point feature extraction network comprises the following steps: carrying out average pooling on the target key point characteristics in the horizontal direction by utilizing at least two pooling kernels with the same width but different lengths; performing linear mapping on the output result of each pooling core by using full-connection layers, wherein the full-connection layers corresponding to different pooling layers are independent of each other; and sequentially arranging and splicing the outputs of the full connection layers according to the sizes of the cores corresponding to the pooled cores.

In order to solve the above technical problem, another technical solution adopted by the present application is: a gait recognition device based on body type transformation is provided, which comprises a memory and a processor, wherein the memory is used for storing program data, and the processor is used for executing the program data to realize the gait recognition method based on body type transformation.

In order to solve the above technical problem, another technical solution adopted by the present application is: there is provided a computer-readable storage medium storing program data for implementing the gait recognition method based on body type transformation as described above when the program data is executed by a processor.

The beneficial effect of this application is: different from the prior art, the gait recognition method based on body type transformation provided by the application obtains a target gait contour map and a target key point matrix of each gait image in a gait image sequence, and carries out aspect ratio transformation on a target contour in the target gait contour map to obtain a target body type transformation contour map; further, acquiring gait space-time characteristics by using a gait space-time extraction network based on a target gait contour map and a target body type transformation contour map, and acquiring target key point characteristics by using a key point characteristic extraction network based on a target key point matrix; and finally, performing feature fusion on the gait space-time features and the target key point features to obtain fusion measurement features. In one embodiment, the combination of the target gait contour map, the target key point matrix and the target body type transformation contour map in each gait image can effectively solve the problem of low gait recognition accuracy caused by clothing shielding, carrying objects and the like, and can improve the accuracy of target gait recognition.

Drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the description of the embodiments are briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present application, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without creative efforts. Wherein:

fig. 1 is a schematic flow chart of a first embodiment of a gait recognition method based on body type transformation provided by the application;

fig. 2 is a schematic flowchart of a second embodiment of a gait recognition method based on body type transformation according to the present application;

FIG. 3 is a schematic flow chart of a third embodiment of a gait recognition method based on body type transformation provided by the application;

FIG. 4 is a schematic flow chart of a fourth embodiment of a gait recognition method based on body type transformation provided by the application;

FIG. 5 is a schematic flow chart of a fifth embodiment of a gait recognition method based on body type transformation adopted by the application;

fig. 6 is a schematic flowchart of a sixth embodiment of a gait recognition method based on body type transformation according to the present application;

FIG. 7 is a schematic structural diagram of an embodiment of a gait recognition device based on body shape transformation provided by the application;

FIG. 8 is a schematic structural diagram of an embodiment of a computer-readable storage medium provided in the present application.

Detailed Description

The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are only a part of the embodiments of the present application, and not all of the embodiments. All other embodiments obtained by a person of ordinary skill in the art based on the embodiments in the present application without making any creative effort belong to the protection scope of the present application.

Referring to fig. 1, fig. 1 is a schematic flow chart of a first embodiment of a gait recognition method based on body type transformation, specifically including the following steps:

step 11: acquiring a gait image sequence, wherein the gait image sequence comprises a plurality of gait images.

Specifically, the gait recognition is similar to the acquisition of human face data, the input is a video sequence, but the gait recognition is to recognize one frame of image, that is, the acquired video data is divided into frames, and each frame of image containing the target person is a gait image.

In one embodiment, the gait images in the sequence of gait images are chronologically arranged.

The step 11 of acquiring the gait image sequence specifically includes the following steps (not shown):

s1: the method comprises the steps of obtaining an original image sequence formed by shooting a target in the walking process, wherein the original image sequence comprises a plurality of original images.

Specifically, a video sequence is segmented by frames based on a segment of the video sequence, and one image is obtained from one frame, that is, the image sequence containing the target can be obtained from the video containing the target.

S2: and respectively carrying out target detection and target tracking on the original image.

Specifically, gait recognition mainly uses gait features to identify a human target, and the gait recognition process can be divided into three parts, namely human target detection, feature extraction and gait recognition. The human body target detection is to detect and extract a specified target from a video or an image sequence, namely comprises two parts of target detection and target tracking.

The target Tracking is classified into single target Tracking (VOT/SOT), multi-Object Tracking (MOT), pedestrian Re-identification (Person Re-ID), multi-target Multi-camera Tracking (MTMCT), pose Tracking, and the like.

The target tracking may track the human target using a tracking algorithm, optionally including depsort, fairMOT, and the like.

Specifically, the deppsort algorithm integrates motion information and appearance information by combining two metrics of Mahalanobis distance (Mahalanobis distance) and feature Cosine distance (Cosine similarity) of the target box. Wherein, the appearance information refers to that a simple CNN network is used to extract the appearance characteristics of the detected object; the motion information refers to the result of kalman (kalman) filtering prediction.

Specifically, fairMOT is an end-to-end algorithm integrating detection and tracking, the detection part is based on CenterNet, and the tracking part is similar to DeepsORT.

S3: selecting the original images with the target leg part not being blocked from the plurality of original images to form the gait image sequence.

Specifically, the target leg is not shielded, which means that no other object is shielded on the human body target in the original image, and the target leg is not shielded by the clothes worn on the target.

As can be appreciated, the gait image sequence consists of several unobstructed images of the target leg.

Step 12: and respectively acquiring a target gait contour map and a target key point matrix corresponding to the same target based on each gait image.

Specifically, in an embodiment, each gait image is input into a trained semantic segmentation model, the gait images are segmented according to frames, the foreground of the gait images is separated from the background, then, binarization processing is performed on the separated foreground image and the background image, that is, the pixel value of the foreground image is set to a first numerical value, the pixel value of the background image is set to a second numerical value different from the first numerical value, and foreground extraction is performed by using a foreground extraction technology. In one embodiment, the pixel value of the foreground map is set to 255 (eight bits) and the pixel value of the background map is set to 0, although in other embodiments, the first pixel value and the second pixel value may be set to other values, such as sixteen bits, twenty-four bits, and so on.

Specifically, in an embodiment, each gait image is input into a trained human body key point detection model, the gait images are detected by frames, and coordinate information and confidence degrees of all key points of a human body target can be obtained, wherein each coordinate information corresponds to one confidence degree, and then the coordinate information of the key points and the confidence degrees of the key points are correspondingly ordered according to a predetermined order (for example, from left to right, from top to bottom, and the like). In one embodiment, there are 17 key points of the human body, including left eye, right eye, nose tip, left ear, right ear, left shoulder, right shoulder, left elbow, right elbow, left wrist, right wrist, left hip, right hip, left knee, right knee, left ankle, and right ankle.

Step 13: and carrying out aspect ratio conversion on the target contour in each target gait contour map to obtain a target body type conversion contour map.

Specifically, a target contour is obtained based on a target gait contour map, the aspect ratio of the target contour is obtained, then the aspect ratio of the target contour is changed to obtain a new target contour, and an image containing the new target contour is a target body type transformation function.

Optionally, the aspect ratio of the new target profile may be a maximum value, a minimum value, a mode, a mean value, a median, and the like of the aspect ratios of all target profiles, or may be an expectation and a standard deviation obtained by normally distributing an aspect ratio set of all target profiles, and then a value range is determined based on the expectation and the standard deviation, for example, a value may be taken from an [ expectation-standard deviation, expectation + standard deviation ] interval.

(

Steps

14 and 15 do not differ in order)

Step 14: and acquiring gait space-time characteristics by utilizing a gait space-time extraction network based on the target gait contour map and the target body type transformation contour map.

In one embodiment, the target gait contour map and the target body shape transformation contour map are binary human body contour maps, and the image sizes are the same (the human body contours are different).

In one embodiment, the Gait space-time extraction Network is an existing Gait recognition Network gap terrestrial Network (GLN). Of course, in other embodiments, there may be other deep gait feature extraction networks including convolutional layers, pooling layers, active layers, full link layers, residual connection and other modules, and the like, which is not limited herein.

Step 15: and acquiring the target key point characteristics by utilizing a key point characteristic extraction network based on the target key point matrix.

Specifically, since the coordinate information of the keypoint in the keypoint matrix corresponds to the confidence corresponding to the coordinate at the spatial position, an m × n matrix may be obtained, where m is the number of the keypoints, and n is equal to the sum of the coordinate dimension and the confidence dimension.

Specifically, in an embodiment, the key point feature extraction network is a graph convolution neural network, and the graph convolution neural network may perform a convolution operation on graph data to extract features of a graph. Since the graph data is data composed of nodes and edges, the graph data can be acquired by using the target key point matrix, wherein human key points can be regarded as nodes, and connection relations among the key points can be regarded as edges for connecting the nodes.

Specifically, the topological characteristic and the time characteristic of the human body key point can be obtained by inputting the target key point matrix into a key point characteristic extraction network.

It is worth noting that the graph convolution neural network adopted in the application is different from the existing graph convolution neural network in that the graph convolution neural network provided in the application is provided with a feature block recombination module which is used for extracting more detailed key point features.

Step 16: and performing feature fusion on the gait space-time feature and the target key point feature to obtain a fusion measurement feature.

Specifically, the gait space-time characteristics and the target key point characteristics are mapped to the same dimension, and characteristic splicing is performed to obtain a plurality of fusion measurement characteristics.

In addition, when the fusion metric features need to be classified, each fusion metric feature can be input into the full link layer and the softmax layer to obtain a fusion classification feature.

In addition, model training is performed by using the fusion measurement features and the fusion classification features, so that a better model (less loss) can be obtained, specifically, a horizontal pyramid method is adopted, the fusion measurement features and the fusion classification features are horizontally partitioned, the horizontal partitions of the fusion measurement features are calculated by using a triple loss function, and the horizontal partitions of the fusion classification features are calculated by using a cross entropy loss function.

In addition, the model can be tested using the fused classification features.

Different from the prior art, the gait recognition method based on body type transformation provided by the application can effectively solve the problems of change of human body contour and the like caused by clothes shielding and carrying objects by acquiring the target gait contour map, the target key point matrix and the target body type transformation contour map in each gait image through combination of the three information, and further improves the accuracy of target gait recognition.

Referring to fig. 2, fig. 2 is a schematic flow chart of a gait recognition method based on body type transformation according to a second embodiment of the present application, and the specific steps are as follows:

step 21: acquiring a gait image sequence, wherein the gait image sequence comprises a plurality of gait images.

Specifically, the gait image sequence is obtained from gait videos, wherein the gait videos include videos of walking of human targets, the gait videos are divided according to frames to obtain a plurality of gait images including the target human body, and the plurality of gait images arranged according to time sequence form the gait image sequence.

Step 22: and respectively acquiring a target gait contour map and a target key point matrix corresponding to the same target based on each gait image.

Specifically, in an embodiment, each gait image is input to a trained semantic segmentation model to obtain a target gait contour map, and the gait images are input to a trained human body key point detection model to obtain a target key point matrix.

(step 23 and step 24 do not exist at the same time)

Step 23: and inputting the original aspect ratio of the target contour of each target gait contour map into a body type transformation function to obtain the corresponding target aspect ratio.

Specifically, the body shape transformation function is determined by the aspect ratio of the target contour in all the gait contour maps.

Specifically, the target aspect ratio may be a mode, a maximum, a minimum, or an average, etc., of the aspect ratios of all target profiles, that is, the target aspect ratio may or may not be the same as the original aspect ratio.

Specifically, the aspect ratio is defined as: acquiring a leftmost point and a rightmost point of the target contour, defining the difference value of the abscissa of the leftmost point and the rightmost point as the width of the target contour, acquiring the highest point and the lowest point of the target contour, defining the difference value of the ordinate of the highest point and the lowest point as the height of the target contour, and determining the difference value of the width of the target contour and the height of the target contour as the aspect ratio of the target contour.

Step 24: obtaining a corresponding target aspect ratio based on expectations and standard deviations of a set of original aspect ratios of a target contour of the target gait contour.

Specifically, in one embodiment, the set of original aspect ratios is normally distributed, i.e., the target aspect ratio is obtained by obtaining the expectation and standard deviation of the normally distributed set of original aspect ratios.

In particular, see fig. 3 for how to obtain the target aspect ratio.

Step 25: and transforming the target contour based on the target width-height ratio to obtain a target body type transformation contour map.

Specifically, based on the target aspect ratio, the target body type transformation contour map can be obtained by performing operations such as stretching and deformation on the target contour frame by frame.

Step 26: and acquiring gait space-time characteristics by utilizing a gait space-time extraction network based on the target gait contour map and the target body type transformation contour map.

Specifically, in an embodiment, the Gait space-time extraction Network is a goal terrestrial Network (GLN), and the features of the time dimension and the space-time dimension can be obtained by transforming the target Gait profile and the target body type transformation profile.

In particular, please refer to fig. 4 for how to obtain gait spatiotemporal characteristics, which will not be described in detail here.

Step 27: and acquiring the target key point characteristics by utilizing a key point characteristic extraction network based on the target key point matrix.

Specifically, in an embodiment, the key point feature extraction network is a graph convolution neural network, and the target key point matrix is input into the graph convolution neural network to obtain the target key point.

Specifically, please refer to fig. 4 for how to obtain the target keypoint features, which will not be described in detail here.

Step 28: and performing feature fusion on the gait space-time feature and the target key point feature to obtain a fusion measurement feature.

Specifically, the gait space-time characteristics and the target key point characteristics are mapped to the same dimension and spliced to obtain fusion measurement characteristics.

Different from the prior art, the gait recognition method based on body type transformation can effectively solve the problem of low gait recognition accuracy caused by clothes shielding, carrying objects and the like, and further improves the accuracy of target gait recognition.

The following is a description of the manner in which the target aspect ratio is obtained.

Referring to fig. 3, fig. 3 is a schematic flow chart of a third embodiment of a gait recognition method based on body type transformation, which specifically includes the following steps:

step 31: calculating one of a maximum, a minimum, a mode, a mean, or a median of the plurality of original aspect ratios as the target aspect ratio using the body type transformation function.

It will be appreciated that the target aspect ratio may or may not be the same as the original aspect ratio.

(Steps 31 and 32-33 are not simultaneous, i.e., the target aspect ratio is obtained in the manner described in step 31 or in the manner described in steps 32-33.)

Step 32: and obtaining the expectation and standard deviation of the normally distributed original aspect ratio set.

It will be appreciated that all of the original aspect ratios are approximated as normal distributions, thereby obtaining the expectation and standard deviation in the normal distributions.

Step 33: the target aspect ratio is selected from a range of values formed by the expectation and standard deviation.

Specifically, in one embodiment, the range of values formed based on the expectation and standard deviation is [ expectation-standard deviation, expectation + standard deviation ].

It should be noted that the target aspect ratio corresponding to each target gait contour map may be the same or different.

Different from the prior art, the method can change the outline of the target in the target gait outline map to obtain the target body type transformation outline map.

The following is a detailed description of the manner in which the spatiotemporal features of gait are obtained.

Referring to fig. 4, fig. 4 is a schematic flowchart of a fourth embodiment of a gait recognition method based on body type transformation, which specifically includes the following steps:

step 41: and splicing the corresponding target gait contour map and the target body type transformation contour map according to the channel dimension to form a fusion contour map.

Specifically, the target gait profile and the target body shape transformation profile are corresponding in size and number, so that the target gait profile and the target body shape transformation profile can be spliced in a channel dimension.

Step 42: and extracting the gait space-time characteristics from the fusion contour map by using the gait space-time extraction network.

Specifically, in an embodiment, the Gait space-time extraction Network is a gate terrestrial Network (GLN), and features of a Gait time dimension and a Gait space dimension can be acquired from the fusion contour map by using the Gait space-time extraction Network.

Different from the prior art, the gait space-time characteristics can be acquired by the method, and the problem of low gait recognition accuracy caused by clothes, carried objects and the like can be effectively solved by the gait space-time characteristics extracted by the method.

The following describes a manner of obtaining the key point features of the target.

Referring to fig. 5, fig. 5 is a schematic flow chart of a fifth embodiment of a gait recognition method based on body type transformation, which specifically includes the following steps:

step 51: and averaging the target key point features in the horizontal direction by using at least two pooling kernels with the same width but different lengths.

Specifically, the pooled kernels used for pooling the target keypoint features are the same size in width but different lengths, e.g., one pooled kernel size is 3*5 and the other pooled kernel size is 5*5.

Step 52: and performing linear mapping on the output result of each pooling core by using a full-connection layer, wherein the full-connection layers corresponding to different pooling layers are independent from each other.

Specifically, a full link layer is connected behind each pooling layer and used for performing linear mapping on the result output by the pooling layer, and the full link layers are independent from each other.

Step 53: and sequentially arranging and splicing the outputs of the full connection layers according to the core size corresponding to the pooled core.

Different from the prior art, the target key point features can be obtained through the method, and the target key point features are recombined key point features.

Referring to fig. 6, fig. 6 is a schematic flowchart of a gait recognition method based on body type transformation according to a sixth embodiment, specifically including the following steps:

step 61: the method comprises the steps of obtaining an original image sequence formed by shooting a target in the walking process, wherein the original image sequence comprises a plurality of original images.

It can be understood that, a video shot by a human body target in the walking process is segmented by frames to obtain a plurality of original images, and each original image contains a human body target.

Step 62: and respectively carrying out target detection and target tracking on the original image.

Specifically, target detection refers to detecting and extracting a specified target from a video or image sequence.

It can be understood that the conventional target detection methods include an optical flow method, a background subtraction method, and the like, and at present, the more popular target detection method is a deep learning motion detection method, which mainly uses a deep learning target detector for detection, and structurally, the deep learning target detector can be divided into a bipolar detector (a candidate area-based detection method) and a single-stage detector (an end-to-end detection method).

And step 63: selecting the original images with the target leg part not being blocked from the plurality of original images to form the gait image sequence.

Specifically, the target leg is not shielded, which means that the human target is not shielded by other objects not attached to the human target.

Step 64: and respectively acquiring a target gait contour map and a target key point matrix corresponding to the same target based on each gait image.

Specifically, a target gait contour map is obtained by inputting a gait image into a trained semantic segmentation model; the target key point matrix is obtained by inputting the gait image into the human body key point detection model.

Step 65: and carrying out aspect ratio conversion on the target contour in each target gait contour map to obtain a target body type conversion contour map.

The aspect ratio transformation of the target contour involves obtaining the aspect ratio of the target, please refer to fig. 3, which is not described herein again.

And step 66: and splicing the corresponding target gait contour map and the target body type transformation contour map according to the channel dimension to form a fusion contour map.

Step 67: and extracting the gait space-time characteristics from the fusion contour map by using the gait space-time extraction network.

Specifically, please refer to fig. 4 for extracting the gait space-time characteristics from the fusion contour by using the gait space-time extraction network, which is not described herein again.

(Steps 66-67 and steps 68-610 do not differ in order)

Step 68: and utilizing at least two pooling kernels with the same width and different lengths to carry out average pooling on the target key point characteristics in the horizontal direction.

Step 69: and performing linear mapping on the output result of each pooling core by using a full connection layer, wherein the full connection layers corresponding to different pooling layers are independent from each other.

Step 610: and sequentially arranging and splicing the outputs of the full connection layers according to the core size corresponding to the pooled core.

Step 611: and performing feature fusion on the gait space-time feature and the target key point feature to obtain a fusion measurement feature.

Different from the prior art, the gait recognition method based on body type transformation can solve the problem of low gait recognition accuracy caused by clothes shielding, carrying objects and the like, and therefore accuracy of target gait recognition is improved.

Referring to fig. 7, fig. 7 is a schematic structural diagram of an embodiment of a gait recognition device based on body type transformation provided in the present application, where the gait recognition device 70 includes a memory 701 and a processor 702, the memory 701 is used for storing program data, and the processor 702 is used for executing the program data to implement the above-mentioned gait recognition method based on body type transformation, which is not described herein again.

Referring to fig. 8, fig. 8 is a schematic structural diagram of an embodiment of a computer-readable storage medium 80 provided in the present application, where the computer-readable storage medium 80 stores program data 801, and when the program data 801 is executed by a processor, the program data 801 is used to implement the gait recognition method based on body type transformation as described above, and details are not repeated here.

The processor of the present application may be referred to as a CPU (Central Processing Unit), an integrated circuit chip, a general-purpose processor, a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field Programmable Gate Array (FPGA) or other programmable logic device, a discrete gate or transistor logic device, or a discrete hardware component. A general purpose processor may be a microprocessor or any conventional processor or the like.

The computer-readable storage medium in the embodiment of the present application may be a medium that can store program instructions, such as a usb disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk, or an optical disk, or may also be a server that stores the program instructions, and the server may send the stored program instructions to other devices for operation, or may self-operate the stored program instructions.

The above description is only for the purpose of illustrating embodiments of the present application and is not intended to limit the scope of the present application, and all modifications of equivalent structures and equivalent processes, which are made by the contents of the specification and the drawings of the present application or are directly or indirectly applied to other related technical fields, are also included in the scope of the present application.

Claims

1. A gait recognition method based on body type transformation is characterized in that the method comprises the following steps:

acquiring a gait image sequence, wherein the gait image sequence comprises a plurality of gait images;

respectively acquiring a target gait contour map and a target key point matrix corresponding to the same target based on each gait image;

carrying out aspect ratio transformation on the target contour in each target gait contour map to obtain a target body type transformation contour map;

acquiring gait space-time characteristics by utilizing a gait space-time extraction network based on the target gait contour map and the target body type transformation contour map;

based on the target key point matrix, extracting a network by using key point features to obtain target key point features;

and performing feature fusion on the gait space-time feature and the target key point feature to obtain a fusion measurement feature.

2. The method of claim 1,

the acquiring of the gait image sequence comprises:

acquiring an original image sequence formed by shooting a target in a walking process, wherein the original image sequence comprises a plurality of original images;

respectively carrying out target detection and target tracking on the original image;

selecting the original images with the target leg part not being blocked from the plurality of original images to form the gait image sequence.

3. The method of claim 1,

the aspect ratio transformation is carried out on the target gait contour map to obtain a target body type transformation contour map, and the method comprises the following steps:

inputting the original aspect ratio of the target contour of each target gait contour map into a body type transformation function to obtain a corresponding target aspect ratio; or

Obtaining a corresponding target aspect ratio based on expectations and standard deviations of a set of original aspect ratios of a target contour of the target gait contour;

and transforming the target contour based on the target width-height ratio.

4. The method of claim 3,

inputting the original aspect ratio of the target contour of each target gait contour map into a body shape transformation function, wherein the method comprises the following steps:

calculating one of a maximum, a minimum, a mode, a mean, or a median of the plurality of original aspect ratios as the target aspect ratio using the body type transformation function.

5. The method of claim 3,

the expectation and standard deviation of the set of original aspect ratios of the target contour based on the target gait contour comprise:

obtaining the expectation and standard deviation of the normally distributed original aspect ratio set;

the target aspect ratio is selected from a range of values formed by the expectation and standard deviation.

6. The method of claim 3 wherein the target aspect ratios for each of the target gait profiles are the same or different.

7. The method of claim 1,

the method for acquiring the gait space-time characteristics by using the gait space-time extraction network based on the target gait contour map and the target body type transformation contour map comprises the following steps:

splicing the corresponding target gait contour map and the target body type transformation contour map according to the channel dimension to form a fusion contour map;

and extracting the gait space-time characteristics from the fusion contour map by using the gait space-time extraction network.

8. The method of claim 1,

the method for acquiring the target key point features by using the key point feature extraction network based on the target key point matrix comprises the following steps:

carrying out average pooling on the target key point characteristics in the horizontal direction by utilizing at least two pooling kernels with the same width but different lengths;

performing linear mapping on the output result of each pooling core by using a full-link layer, wherein the full-link layers corresponding to different pooling layers are independent of each other;

and sequentially arranging and splicing the outputs of the full connection layers according to the core size corresponding to the pooled cores.

9. A gait recognition device based on body type transformation, characterized in that the device comprises a memory storing program data and a processor for executing the program data to implement the gait recognition method according to any one of claims 1 to 8.

10. A computer-readable storage medium, in which program data are stored which, when being executed by a processor, are adapted to carry out the method of any one of claims 1-8.