Pedestrian gait recognition method based on neural network

Abstract

The invention discloses a pedestrian gait recognition method based on a neural network, and the method has the beneficial effects that a pedestrian gait image is processed through the neural network and a deep learning technology, pedestrian gait silhouette features are obtained, the pedestrian gait silhouette features are preprocessed, and a preprocessed pedestrian gait silhouette image is obtained; the pedestrian gait silhouette image is sent to a target CNN neural network model for learning processing, and pedestrian gait features are obtained; inputting the pedestrian gait features into a target pooling module for special pooling processing to obtain pooled pedestrian gait features; inputting the pooling pedestrian gait features into a full-link model to obtain a full-link layer; inputting the full-link layer into a gait recognition model, and outputting a gait recognition result; and if the gait recognition result is matched with the pedestrian identity in the database, outputting the pedestrian identity information. Therefore, traffic illegal behaviors such as red light running and road crossing of pedestrians can be monitored, so that the traffic safety is improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent detection technology, and in particular to a pedestrian gait recognition method based on a neural network. Background Art

[0002] Gait recognition is one of the most important and effective biometric recognition technologies, which can authenticate identity away from the camera without the cooperation of the subject. Gait recognition has a wide range of applications in social security and identity recognition. As a recognition task in vision, the basic goal of gait recognition is to learn a unique and invariant representation from the temporal changes of human body shape. The research on gait recognition technology has important application value. Summary of the invention

[0003] The purpose of the present invention is to solve the above problems and designs a pedestrian gait recognition method based on neural network.

[0004] To achieve the above object, the technical solution of the present invention is that, further, in the above-mentioned pedestrian gait recognition method based on neural network, the pedestrian gait recognition method comprises the following steps:

[0005] Processing pedestrian gait images using neural networks and deep learning technology to obtain pedestrian gait silhouette features, preprocessing the pedestrian gait silhouette features to obtain preprocessed pedestrian gait silhouette images;

[0006] The pedestrian gait silhouette image is sent into the target CNN neural network model for learning processing to obtain pedestrian gait features; the pedestrian gait features are input into the target pooling module for special pooling processing to obtain pooled pedestrian gait features; the pooled pedestrian gait features are input into the full-link model to obtain a full-link layer;

[0007] Input the fully connected layer into the gait recognition model and output the gait recognition result;

[0008] If the gait recognition result matches the pedestrian identity in the database, the pedestrian identity information is output.

[0009] Furthermore, in the above-mentioned pedestrian gait recognition method based on neural network, the pedestrian gait image is processed by using neural network and deep learning technology to obtain pedestrian gait silhouette features, and the pedestrian gait silhouette features are preprocessed to obtain the preprocessed pedestrian gait silhouette image, including:

[0010] Acquire pedestrian gait image data, perform data cleaning on the pedestrian gait image data, remove invalid samples, and obtain cleaned pedestrian gait image data;

[0011] The cleaned pedestrian gait image data is converted into a binary image using an opencv library, and data enhancement is performed to obtain enhanced pedestrian gait image data;

[0012] The enhanced pedestrian gait image data is scaled by using maximum and minimum normalization to obtain pre-processed pedestrian gait image data of a size of 64×64.

[0013] Further, in the above-mentioned pedestrian gait recognition method based on neural network, the pedestrian gait silhouette image is sent to the target CNN neural network model for learning processing to obtain pedestrian gait features; the pedestrian gait features are input into the target pooling module for special pooling processing to obtain pooled pedestrian gait features; the pooled pedestrian gait features are input into the full-link model to obtain the full-link layer, including:

[0014] A ResNet-like neural network based on an FRN layer and a TLU activation function is used to extract the pedestrian gait image data using a plurality of multi-channel two-dimensional convolutional filter variants with a size of 3×3 to obtain spatial features and temporal features of the feature map sequence;

[0015] A pooling module based on a special pooling function of average pooling and variance pooling fitting is used to aggregate the spatial and temporal features of the feature map sequence using multiple multi-channel two-dimensional convolutional filter variants with a size of 3×3 along the channel dimension;

[0016] The spatial features and temporal features of the feature map sequence are fused through horizontal pyramid pooling, each feature map is horizontally divided into 8 even parts, and input into the fully connected model to obtain the fully connected layer.

[0017] Furthermore, in the above-mentioned pedestrian gait recognition method based on a neural network, the ResNet-like neural network using the FRN layer and the TLU activation function as the main body uses multiple multi-channel two-dimensional convolution filter variants with a size of 3×3 to extract the pedestrian gait image data, and obtains the spatial features and temporal features of the feature map sequence, including:

[0018] In the ResNet-like neural network, the FRN layer and TLU activation function are used to replace the original BN layer and ReLu activation function;

[0019] The number of channels of the feature map in each layer of the neural network is obtained, the number of the multi-channel two-dimensional convolution filters is set to the number of channels of the feature map sequence, and the spatial features and temporal features after convolution are obtained.

[0020] Furthermore, in the above-mentioned pedestrian gait recognition method based on a neural network, the ResNet-like neural network using the FRN layer and the TLU activation function as the main body uses multiple multi-channel two-dimensional convolution filter variants with a size of 3×3 to extract the pedestrian gait image data to obtain the spatial features and temporal features of the feature map sequence, and also includes:

[0021] Obtaining a feature map of the neural network input, processing the feature map using a plurality of variance and mean pooling filter variants with a size of 3×3, and obtaining a pooled pedestrian gait feature under a certain channel;

[0022] The number of channels of the feature map sequence is obtained, and the number of multi-channel variance and average pooling filter variants is set to the number of channels of the feature map sequence to obtain the spatial features and temporal features of the pedestrian gait features after pooling.

[0023] Furthermore, in the above-mentioned pedestrian gait recognition method based on a neural network, the spatial features and temporal features of the feature map sequence are fused by horizontal pyramid pooling, each feature map is horizontally divided into 8 even parts, and input into the full-link model to obtain a full-link layer, including:

[0024] The spatial features and temporal features of the pedestrian gait features after pooling are obtained, the parameter of the horizontal pyramid pooling is set to 8, and the number of the horizontal pyramid pooling is set to the number of channels of the feature map sequence, so as to obtain a fully connected layer with a channel number of 1.

[0025] Furthermore, in the above-mentioned pedestrian gait recognition method based on a neural network, the step of inputting the full-link layer into the gait recognition model and outputting the gait recognition result includes:

[0026] Obtain a fully connected layer with a channel number of 1, input the fully connected layer into a gait recognition model, perform Euclidean distance discrimination on the fully connected layer through the gait recognition model, and generate a Euclidean distance;

[0027] Acquire gait data in the database, input the gait data in the database into a gait recognition model, perform Euclidean distance discrimination on the full-link layer through the gait recognition model, and generate Euclidean distance;

[0028] The Euclidean distance of the full link layer and the Euclidean distance of the gait data in the database are obtained, the Euclidean length between the two is calculated, and a score is calculated according to the difference in length, and a gait recognition result is output based on the score.

[0029] Further, in the above-mentioned pedestrian gait recognition method based on a neural network, if the gait recognition result matches the pedestrian identity in the database, outputting the pedestrian identity information includes:

[0030] When the Euclidean distance of the full link layer and the Euclidean distance of the gait data in the database are calculated, the calculation score is determined, and the ranking is performed according to the score. The output result is the pedestrian identity information in the first corresponding database, that is, the pedestrian identity information in the corresponding database whose gait data is most similar to the two.

[0031] Furthermore, in a system for implementing the above-mentioned pedestrian gait recognition method based on a neural network, the system includes a preprocessing module, a feature extraction module, a gait recognition module and an output information module, wherein:

[0032] The preprocessing module uses neural network and deep learning technology to process the pedestrian gait image to obtain the pedestrian gait silhouette features, and preprocesses the pedestrian gait silhouette features to obtain the preprocessed pedestrian gait silhouette image;

[0033] The feature extraction module is used to process the spatial and temporal features in the preprocessed gait silhouette to obtain a gait feature graph sequence, perform feature extraction on the gait feature graph sequence to obtain pedestrian gait features; perform pooling and full-link operations on the pedestrian gait features to obtain a gait feature full-link layer.

[0034] A gait recognition module, used for inputting the gait feature full-link layer into a gait recognition model, and outputting a gait recognition result by calculating the Euclidean distance thereof;

[0035] The output information module is used to determine the calculation score after the Euclidean distance of the full link layer and the Euclidean distance of the gait data in the database are calculated, and rank them according to the scores. The output result is the pedestrian identity information in the most similar corresponding database.

[0036] Furthermore, in a system for implementing the above-mentioned pedestrian gait recognition method based on a neural network, the system includes a feature extraction module, a special pooling module, a horizontal pyramid pooling module, and a full-link module, wherein:

[0037] The feature extraction module uses the FRN layer and TLU activation function to replace the original BN layer and ReLu activation function in the ResNet-like neural network to extract the features of the 64×64 gait image and obtain the spatial and temporal features of the gait sequence;

[0038] A special pooling module uses a pooling function that is fitted by variance pooling plus mean pooling to further condense and compress the spatial and temporal features of the gait sequence to obtain pooled gait sequence features;

[0039] A horizontal pyramid pooling module is used to horizontally divide the gait sequence features into 8 parts to obtain 8 equally divided feature strips;

[0040] The full-link module is used for the 8-equal feature strips after full-linking, and the full-linking operation is adopted to obtain the full-linking layer.

[0041] Its beneficial effect is that by using neural network and deep learning technology to process pedestrian gait images, pedestrian gait silhouette features are obtained, the pedestrian gait silhouette features are preprocessed to obtain the preprocessed pedestrian gait silhouette image; the pedestrian gait silhouette image is sent to the target CNN neural network model for learning and processing to obtain pedestrian gait features; the pedestrian gait features are input into the target pooling module for special pooling processing to obtain pooled pedestrian gait features; the pooled pedestrian gait features are input into the full-link model to obtain the full-link layer; the full-link layer is input into the gait recognition model to output the gait recognition result; if the gait recognition result matches the pedestrian identity in the database, the pedestrian identity information is output. First of all, in the field of public security, through pedestrian gait recognition technology, it is possible to achieve close monitoring of key areas such as subway stations, central streets and hospitals where there are dense crowds of people, and to timely warn and handle abnormal situations, thereby ensuring the safety of people's lives and property and the stability of social security. Secondly, in the field of smart business, by tracking and analyzing the whereabouts of consumers, personalized services and product recommendations can be achieved, improving customer satisfaction and store sales, etc. In the field of smart transportation, pedestrian gait recognition technology can be used to monitor traffic violations, such as pedestrians running red lights and crossing the road, thereby improving traffic safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the following detailed description of the preferred embodiment.The drawings are only for the purpose of illustrating the preferred embodiments and are not to be construed as limiting the invention.

[0043] Figure 1 Schematic diagram of a first embodiment of a pedestrian gait recognition method based on a neural network in an embodiment of the present invention;

[0044] Figure 2 Schematic diagram of a second embodiment of a pedestrian gait recognition method based on a neural network in an embodiment of the present invention. DETAILED DESCRIPTION

[0045] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0046] Those skilled in the art will appreciate that, unless otherwise stated, the singular forms "a", "an", "said" and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present invention refers to the presence of the features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof.

[0047] The present invention will be described in detail below in conjunction with the accompanying drawings. Figure 1 As shown, a pedestrian gait recognition method based on a neural network comprises the following steps:

[0048] Step 101: Process the pedestrian gait image using a neural network and deep learning technology to obtain pedestrian gait silhouette features, and pre-process the pedestrian gait silhouette features to obtain a pre-processed pedestrian gait silhouette image;

[0049] Specifically, in this embodiment, pedestrian gait image data is obtained, and data cleaning is performed on the pedestrian gait image data to remove invalid samples to obtain cleaned pedestrian gait image data; the cleaned pedestrian gait image data is converted into a binary image using the opencv library, and data enhancement is performed to obtain enhanced pedestrian gait image data; the enhanced pedestrian gait image data is scaled using maximum and minimum normalization to obtain preprocessed pedestrian gait image data of size 64×64.

[0050] Step 102: Send the pedestrian gait silhouette image into the target CNN neural network model for learning processing to obtain the pedestrian gait features; input the pedestrian gait features into the target pooling module for special pooling processing to obtain the pooled pedestrian gait features; input the pooled pedestrian gait features into the full-link model to obtain the full-link layer;

[0051] Specifically, in this embodiment, a ResNet-like neural network with an FRN layer and a TLU activation function as the main body is used, and multiple multi-channel two-dimensional convolution filter variants with a size of 3×3 are used to extract pedestrian gait image data to obtain spatial features and temporal features of a feature map sequence;

[0052] A pooling module based on a special pooling function fitted by average pooling and variance pooling. Along the channel dimension, multiple multi-channel two-dimensional convolutional filter variants with a size of 3×3 are used to aggregate the spatial and temporal features of the feature map sequence.

[0053] The spatial and temporal features of the feature map sequence are fused through horizontal pyramid pooling, each feature map is horizontally divided into 8 even parts, and input into the fully connected model to obtain the fully connected layer.

[0054] Using a ResNet-like neural network with FRN layer and TLU activation function as the main body, multiple multi-channel two-dimensional convolution filter variants with a size of 3×3 are used to extract pedestrian gait image data, and the spatial and temporal features of the feature map sequence are obtained, including: in the ResNet-like neural network, the original BN layer and ReLu activation function are replaced by FRN layer and TLU activation function; the number of channels of the feature map in each layer of the neural network is obtained, and the number of multi-channel two-dimensional convolution filters is set to the number of channels of the feature map sequence to obtain the spatial and temporal features after convolution. The feature map input by the neural network is obtained, and the feature map is processed by multiple variance and average pooling filter variants with a size of 3×3 to obtain the gait features of the pedestrian after pooling under a certain channel; the number of channels of the feature map sequence is obtained, and the number of multi-channel variance and average pooling filter variants is set to the number of channels of the feature map sequence to obtain the spatial and temporal features of the pedestrian gait features after pooling.

[0055] The spatial and temporal features of the pedestrian gait features after pooling are obtained, the parameter of the horizontal pyramid pooling is set to 8, and the number of horizontal pyramid pooling is set to the number of channels of the feature map sequence, and a fully connected layer with a channel number of 1 is obtained.

[0056] Step 103: input the full-link layer into the gait recognition model and output the gait recognition result;

[0057] Specifically, in this embodiment, a full-link layer with a channel number of 1 is obtained, the full-link layer is input into the gait recognition model, and the Euclidean distance is judged on the full-link layer through the gait recognition model to generate the Euclidean distance;

[0058] Obtain gait data in the database, input the gait data in the database into the gait recognition model, perform Euclidean distance judgment on the full connection layer through the gait recognition model, and generate Euclidean distance;

[0059] Get the Euclidean distance of the full link layer and the Euclidean distance of the gait data in the database, calculate the Euclidean length between the two, calculate the score based on the difference in length, and output the gait recognition result based on the score.

[0060] Step 104: If the gait recognition result matches the pedestrian identity in the database, the pedestrian identity information is output.

[0061] Specifically, in this embodiment, when the Euclidean distance of the full-link layer and the Euclidean distance of the gait data in the database are calculated, the calculated scores are determined, and the scores are ranked, and the output result is the pedestrian identity information in the first corresponding database, that is, the pedestrian identity information in the corresponding database whose gait data is most similar to the two.

[0062] Its beneficial effect is that by using neural network and deep learning technology to process pedestrian gait images, pedestrian gait silhouette features are obtained, the pedestrian gait silhouette features are preprocessed to obtain the preprocessed pedestrian gait silhouette image; the pedestrian gait silhouette image is sent to the target CNN neural network model for learning and processing to obtain pedestrian gait features; the pedestrian gait features are input into the target pooling module for special pooling processing to obtain pooled pedestrian gait features; the pooled pedestrian gait features are input into the full-link model to obtain the full-link layer; the full-link layer is input into the gait recognition model to output the gait recognition result; if the gait recognition result matches the pedestrian identity in the database, the pedestrian identity information is output. First of all, in the field of public security, through pedestrian gait recognition technology, it is possible to achieve close monitoring of key areas such as subway stations, central streets and hospitals where there are dense crowds of people, and to timely warn and handle abnormal situations, thereby ensuring the safety of people's lives and property and the stability of social security. Secondly, in the field of smart business, by tracking and analyzing the whereabouts of consumers, personalized services and product recommendations can be achieved, improving customer satisfaction and store sales, etc. In the field of smart transportation, pedestrian gait recognition technology can be used to monitor traffic violations, such as pedestrians running red lights and crossing the road, thereby improving traffic safety.

[0063] like Figure 2 As shown, in a pedestrian gait recognition method based on a neural network of the present invention, the full-link layer is input into a gait recognition model, and the output gait recognition result includes.

[0064] Step 201: obtain a fully connected layer with a channel number of 1, input the fully connected layer into a gait recognition model, perform Euclidean distance judgment on the fully connected layer through the gait recognition model, and generate a Euclidean distance;

[0065] Step 202: Obtain gait data in the database, input the gait data in the database into a gait recognition model, perform Euclidean distance discrimination on the full-link layer through the gait recognition model, and generate Euclidean distance;

[0066] Step 203: Obtain the Euclidean distance of the full link layer and the Euclidean distance of the gait data in the database, calculate the Euclidean length between the two, calculate a score based on the length difference, and output a gait recognition result based on the score.

[0067] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. The above embodiments and descriptions are only preferred examples of the present invention and are not intended to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A pedestrian gait recognition method based on neural network, characterized in that: The pedestrian gait recognition method comprises the following steps: Processing pedestrian gait images using neural networks and deep learning technology to obtain pedestrian gait silhouette features, preprocessing the pedestrian gait silhouette features to obtain preprocessed pedestrian gait silhouette images; The pedestrian gait silhouette image is sent into the target CNN neural network model for learning processing to obtain pedestrian gait features; the pedestrian gait features are input into the target pooling module for special pooling processing to obtain pooled pedestrian gait features; the pooled pedestrian gait features are input into the full-link model to obtain a full-link layer; Input the fully connected layer into the gait recognition model and output the gait recognition result; If the gait recognition result matches the pedestrian identity in the database, the pedestrian identity information is output.

2. A method for pedestrian gait recognition based on a neural network as claimed in claim 1, characterized in that: The method of processing the pedestrian gait image by using a neural network and deep learning technology to obtain a pedestrian gait silhouette feature, and preprocessing the pedestrian gait silhouette feature to obtain a preprocessed pedestrian gait silhouette image includes: Acquire pedestrian gait image data, perform data cleaning on the pedestrian gait image data, remove invalid samples, and obtain cleaned pedestrian gait image data; The cleaned pedestrian gait image data is converted into a binary image using an opencv library, and data enhancement is performed to obtain enhanced pedestrian gait image data; The enhanced pedestrian gait image data is scaled by using maximum and minimum normalization to obtain pre-processed pedestrian gait image data of a size of 64×64.

3. A method for pedestrian gait recognition based on a neural network as claimed in claim 1, characterized in that: The pedestrian gait silhouette image is sent into the target CNN neural network model for learning processing to obtain the pedestrian gait features; the pedestrian gait features are input into the target pooling module for special pooling processing to obtain the pooled pedestrian gait features; The pooled pedestrian gait features are input into the fully connected model to obtain a fully connected layer, including: A ResNet-like neural network based on an FRN layer and a TLU activation function is used to extract the pedestrian gait image data using a plurality of multi-channel two-dimensional convolutional filter variants with a size of 3×3 to obtain spatial features and temporal features of the feature map sequence; A pooling module based on a special pooling function of average pooling and variance pooling fitting is used to aggregate the spatial and temporal features of the feature map sequence using multiple multi-channel two-dimensional convolutional filter variants with a size of 3×3 along the channel dimension; The spatial features and temporal features of the feature map sequence are fused through horizontal pyramid pooling, each feature map is horizontally divided into 8 even parts, and input into the fully connected model to obtain the fully connected layer.

4. A method for pedestrian gait recognition based on a neural network as claimed in claim 3, characterized in that: The ResNet-like neural network using the FRN layer and the TLU activation function as the main body uses multiple multi-channel two-dimensional convolution filter variants with a size of 3×3 to extract the pedestrian gait image data to obtain the spatial features and temporal features of the feature map sequence, including: In the ResNet-like neural network, the FRN layer and TLU activation function are used to replace the original BN layer and ReLu activation function; The number of channels of the feature map in each layer of the neural network is obtained, the number of the multi-channel two-dimensional convolution filters is set to the number of channels of the feature map sequence, and the spatial features and temporal features after convolution are obtained.

5. The method for pedestrian gait recognition based on a neural network as claimed in claim 3, characterized in that: The ResNet-like neural network using the FRN layer and the TLU activation function as the main body, using multiple multi-channel two-dimensional convolution filter variants with a size of 3×3 to extract the pedestrian gait image data, and obtain the spatial features and temporal features of the feature map sequence, also includes: Obtaining a feature map of the neural network input, processing the feature map using a plurality of variance and mean pooling filter variants with a size of 3×3, and obtaining a pooled pedestrian gait feature under a certain channel; The number of channels of the feature map sequence is obtained, and the number of multi-channel variance and average pooling filter variants is set to the number of channels of the feature map sequence to obtain the spatial features and temporal features of the pedestrian gait features after pooling.

6. A method for pedestrian gait recognition based on a neural network as claimed in claim 3, characterized in that: The spatial features and temporal features of the feature map sequence are fused by horizontal pyramid pooling, each feature map is horizontally divided into 8 even parts, and input into the full-link model to obtain a full-link layer, including: The spatial features and temporal features of the pedestrian gait features after pooling are obtained, the parameter of the horizontal pyramid pooling is set to 8, and the number of the horizontal pyramid pooling is set to the number of channels of the feature map sequence, so as to obtain a fully connected layer with a channel number of 1.

7. The method for pedestrian gait recognition based on a neural network as claimed in claim 1, characterized in that: The step of inputting the full-link layer into the gait recognition model and outputting the gait recognition result includes: Obtain a fully connected layer with a channel number of 1, input the fully connected layer into a gait recognition model, perform Euclidean distance discrimination on the fully connected layer through the gait recognition model, and generate a Euclidean distance; Acquire gait data in the database, input the gait data in the database into a gait recognition model, perform Euclidean distance discrimination on the full-link layer through the gait recognition model, and generate Euclidean distance; The Euclidean distance of the full link layer and the Euclidean distance of the gait data in the database are obtained, the Euclidean length between the two is calculated, and a score is calculated according to the difference in length, and a gait recognition result is output based on the score.

8. The method for pedestrian gait recognition based on a neural network as claimed in claim 1, characterized in that: If the gait recognition result matches the pedestrian identity in the database, outputting the pedestrian identity information includes: When the Euclidean distance of the full link layer and the Euclidean distance of the gait data in the database are calculated, the calculation score is determined, and the ranking is performed according to the score. The output result is the pedestrian identity information in the first corresponding database, that is, the pedestrian identity information in the corresponding database whose gait data is most similar to the two.

9. A system for implementing the method for pedestrian gait recognition based on a neural network as claimed in claim 1, characterized in that: The system includes a preprocessing module, a feature extraction module, a gait recognition module and an output information module, wherein: The preprocessing module uses neural network and deep learning technology to process the pedestrian gait image to obtain the pedestrian gait silhouette features, and preprocesses the pedestrian gait silhouette features to obtain the preprocessed pedestrian gait silhouette image; The feature extraction module is used to process the spatial and temporal features in the preprocessed gait silhouette to obtain a gait feature graph sequence, perform feature extraction on the gait feature graph sequence to obtain pedestrian gait features; perform pooling and full-link operations on the pedestrian gait features to obtain a gait feature full-link layer. A gait recognition module, used for inputting the gait feature full-link layer into a gait recognition model, and outputting a gait recognition result by calculating the Euclidean distance thereof; The output information module is used to determine the calculation score after the Euclidean distance of the full link layer and the Euclidean distance of the gait data in the database are calculated, and rank them according to the scores. The output result is the pedestrian identity information in the most similar corresponding database.

10. A system for implementing the method for pedestrian gait recognition based on a neural network as claimed in claim 1, characterized in that: The system includes a feature extraction module, a special pooling module, a horizontal pyramid pooling module, and a full link module, wherein: The feature extraction module uses the FRN layer and TLU activation function to replace the original BN layer and ReLu activation function in the ResNet-like neural network to extract the features of the 64×64 gait image and obtain the spatial and temporal features of the gait sequence; A special pooling module uses a pooling function that is fitted by variance pooling plus mean pooling to further condense and compress the spatial and temporal features of the gait sequence to obtain pooled gait sequence features; A horizontal pyramid pooling module is used to horizontally divide the gait sequence features into 8 parts to obtain 8 equally divided feature strips; The full-link module is used for the 8-equal feature strips after full-linking, and the full-linking operation is adopted to obtain the full-linking layer.

Images