A gait recognition improvement method, system, device and medium

Abstract

This application discloses an improved gait recognition method, comprising: constructing a passenger natural gait dataset for railway station entry; constructing an improved gait recognition network model based on GaitSet, the improved gait recognition network model comprising: a multi-scale feature fusion module and a key feature enhancement module, wherein the multi-scale feature fusion module is used to fuse shallow detail feature information and deep semantic feature information of convolutional neural networks; the key feature enhancement module is used to enhance key features of gait information; and training the improved gait recognition network model using the passenger natural gait dataset, the CASIA-B gait dataset, and the GREW gait dataset to verify the recognition rate of the improved gait recognition network model. This application also discloses an improved gait recognition system.

Description

Technical Field

[0001] This application relates to a gait recognition method, and more particularly to an improved GaitSet gait recognition method and system thereof. Background Technology

[0002] Currently, significant progress has been made in gait recognition algorithms, with recognition accuracy exceeding 95% under laboratory conditions. However, in real-world open environments, gait sequences are exceptionally complex, often involving pedestrian occlusion and pauses, leading to a significant performance degradation in laboratory settings. Most high-accuracy gait recognition algorithms rely on temporal information, thus lacking the ability to handle pedestrian pauses. In this context, the GaitSet algorithm stands out by treating the gait sequence as an unordered set, independent of its temporal information. Therefore, this invention selects the GaitSet algorithm as the baseline for gait recognition tasks in open environments. However, the GaitSet baseline algorithm's feature extraction network is relatively simple and does not fully utilize gait data information.

[0003] Therefore, there is an urgent need to propose an improved gait recognition method and system. A gait recognition algorithm based on multi-scale feature fusion and key feature enhancement is adopted. The algorithm is verified in a dataset in the open environment of the railway passenger service industry and achieves better results than the benchmark algorithm. It solves the problem that the existing technology is not strong enough in solving the problem of pedestrian pauses. In addition, the feature extraction network of the GaitSet benchmark algorithm is relatively simple and does not make full use of gait data information. Summary of the Invention

[0004] This application provides an improved gait recognition method and system, which solves the problems of insufficient ability to solve the problem of pedestrian pauses in the prior art, and the fact that the feature extraction network of the GaitSet benchmark algorithm is relatively simple and does not make full use of gait data information.

[0005] In a first aspect, embodiments of this application provide an improved gait recognition method, the method comprising:

[0006] Steps for constructing the sample set: Construct a dataset of passengers' natural gait during the railway station entry process;

[0007] The steps for constructing an improved gait recognition network model based on GaitSet are as follows: The improved gait recognition network model based on GaitSet includes a multi-scale feature fusion module and a key feature enhancement module. The multi-scale feature fusion module is used to fuse shallow detail feature information and deep semantic feature information from the convolutional neural network; the key feature enhancement module is used to enhance the key features of gait information.

[0008] Gait recognition model training steps: Using the passenger natural gait dataset, CASIA-B gait dataset and GREW gait dataset, train the improved gait recognition network model and verify the recognition rate of the improved gait recognition network model.

[0009] In a specific embodiment of the present invention, the step of constructing the gait recognition network based on GaitSet further includes:

[0010] Feature acquisition steps: Input the gait sequence into the first convolutional neural network and pooling layer of the benchmark model to obtain the first layer of features; input the first layer of features into the second convolutional neural network and pooling layer of the benchmark model to obtain the second layer of features;

[0011] Multi-scale feature fusion steps: Input the first layer features into the multi-scale feature fusion module, perform downsampling to obtain the first sampled value, fuse it with the second layer features to obtain the first fused feature, which is used as the input and relearning of the subsequent neural network.

[0012] In a specific embodiment of the present invention, the step of constructing the gait recognition network based on GaitSet further includes:

[0013] The re-fusion step is as follows: The first fused features are input into the third layer of the convolutional neural network of the baseline model to obtain the third layer features; the first fused features are input into the multi-scale feature fusion module to perform downsampling to obtain the second sampled values, which are then fused with the third layer features to obtain the second fused features.

[0014] In a specific embodiment of the present invention, the step of constructing the gait recognition network based on GaitSet further includes:

[0015] Key feature enhancement steps: After performing pooling on the second fused features to form the fourth layer features, the fourth layer features are input into the key feature enhancement module for feature enhancement; the fourth layer features are fused with the output of MGP and then input into the key feature enhancement module for feature enhancement.

[0016] The key feature enhancement module enhances the feature details of the gait sequence at the pixel level through a spatial attention mechanism, and outputs the enhanced features to perform subsequent operations.

[0017] In this embodiment of the invention, the above-mentioned step of constructing a sample set further includes:

[0018] Collect and create the railway industry gait recognition dataset RENG.

[0019] In this embodiment of the invention, the above-mentioned key feature enhancement step further includes:

[0020] Initialization steps: Perform max pooling and average pooling operations on the input augmented features in the channel dimension to obtain feature maps of size 1×h×w, where 1 represents the number of channels, and h and w are the height and width, respectively.

[0021] Channel stacking step: Stack the 1×h×w feature maps in the channel dimension, input them into a convolutional network for fusion learning, and obtain a feature map of size 1×h×w. Then, perform a Sigmoid activation operation and multiply it with the original feature map to obtain the enhanced feature.

[0022] In this embodiment of the invention, the above-mentioned multi-scale fusion feature f i+1 :

[0023] f i+1 =B(f i )+f i , where f i For the features of the current layer, f i+1 f i The next layer of features, B is a convolutional layer.

[0024] Secondly, embodiments of this application provide an improved gait recognition system, employing the above-described improved gait recognition method. The system includes:

[0025] Sample set construction module: Construct a dataset of passengers' natural gait during the railway station entry process;

[0026] A module for constructing an improved gait recognition network model based on GaitSet is included. This improved gait recognition network model includes a multi-scale feature fusion module and a key feature enhancement module. The multi-scale feature fusion module is used to fuse shallow detail feature information and deep semantic feature information from the convolutional neural network. The key feature enhancement module is used to enhance the key features of gait information.

[0027] Gait recognition model training module: Using the passenger natural gait dataset, CASIA-B gait dataset and GREW gait dataset, an improved gait recognition network model is trained, and the recognition rate of the improved gait recognition network model is verified.

[0028] Thirdly, embodiments of this application provide a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the gait recognition improvement method as described above.

[0029] Fourthly, embodiments of this application provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the gait recognition improvement method as described above.

[0030] Compared with existing technologies, it has the following outstanding advantages:

[0031] 1) The method of this invention proposes a multi-scale feature fusion method, which fuses detailed information and semantic information to obtain more discriminative features;

[0032] 2) The method of this invention proposes a key feature enhancement method, which mines and enhances key features of gait information through an attention mechanism to improve the differences between different gait modes;

[0033] 3) The method of this invention collected and created a gait dataset in an open environment, and the proposed method achieved better results on the dataset of this invention;

[0034] 4) Compared with other methods that simply fuse features from each stage as the final feature, the multi-scale feature fusion method of this invention also passes the fused features to subsequent convolutional layers for further learning, which can better learn details and semantic information and further alleviate gradient vanishing. Attached Figure Description

[0035] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0036] Figure 1 This is a schematic diagram of the improved gait recognition method of the present invention;

[0037] Figure 2 This is a schematic diagram of the gait recognition network structure according to a specific embodiment of the present invention;

[0038] Figure 3 This is a schematic diagram of the FEM module in a specific embodiment of the present invention;

[0039] Figure 4 This is a schematic diagram of the spatial attention mechanism in a specific embodiment of the present invention;

[0040] Figure 5 This is a schematic diagram illustrating a specific embodiment of the RENG dataset of the present invention;

[0041] Figure 6 This is a schematic diagram of the improved gait recognition system of the present invention;

[0042] Figure 7 This is a schematic diagram of the hardware structure of a computer device according to an embodiment of this application.

[0043] In the above image:

[0044] 10. Sample set construction module; 20. Improved gait recognition network model based on GaitSet module.

[0045] 30 Gait Recognition Model Training Module. Detailed Implementation

[0046] To make the objectives, technical solutions, and advantages of this application clearer, the application is described and illustrated below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application. All other embodiments obtained by those skilled in the art based on the embodiments provided in this application without inventive effort are within the scope of protection of this application.

[0047] Obviously, the accompanying drawings described below are merely some examples or embodiments of this application. Those skilled in the art can apply this application to other similar scenarios based on these drawings without any inventive effort. Furthermore, it is understood that although the efforts made in this development process may be complex and lengthy, for those skilled in the art related to the content disclosed in this application, any changes to design, manufacturing, or production based on the technical content disclosed in this application are merely conventional technical means and should not be construed as insufficient disclosure of the content of this application.

[0048] Details of one or more embodiments of this application are set forth in the following drawings and description to make other features, objects and advantages of this application more readily apparent.

[0049] In this application, the reference to "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment that is mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described in this application may be combined with other embodiments without conflict.

[0050] Unless otherwise defined, the technical or scientific terms used in this application shall have the ordinary meaning understood by one of ordinary skill in the art to which this application pertains. The terms “a,” “an,” “an,” “the,” and similar words used in this application do not indicate quantity limitation and may indicate singular or plural. The terms “comprising,” “including,” “having,” and any variations thereof used in this application are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or modules (units) is not limited to the listed steps or units, but may also include steps or units not listed, or may include other steps or units inherent to these processes, methods, products, or devices. The terms “connected,” “linked,” “coupled,” and similar words used in this application are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. “Multiple” used in this application refers to two or more. “And / or” describes the relationship between related objects, indicating that three relationships may exist; for example, “A and / or B” can represent: A alone, A and B simultaneously, and B alone. The character " / " generally indicates that the preceding and following objects are in an "or" relationship. The terms "first," "second," and "third" used in this application are merely to distinguish similar objects and do not represent a specific ordering of the objects.

[0051] This invention aims to provide a gait recognition algorithm based on multi-scale feature fusion and key feature enhancement. The algorithm is based on the GaitSet concept and does not require the gait sequence to be temporally continuous. Because features from different layers of a convolutional neural network have different characteristics—shallow features mainly contain a large amount of detailed information, while deep features contain rich semantic information—combining shallow and deep features and feeding them into the neural network for learning can yield more discriminative features. Furthermore, the differences in gait features between different individuals can sometimes be very small. Therefore, enhancing the key features of gait information can, to some extent, differentiate between different gaits and improve the algorithm's accuracy. Therefore, the gait recognition algorithm based on multi-scale feature fusion and key feature enhancement proposed in this invention is applicable to passenger gait in natural scenarios such as railway station entry. The advantages of this invention are:

[0052] 1. A novel deep learning network model based on the GaitSet gait recognition algorithm is proposed, and a new gait recognition algorithm based on multi-scale fusion and feature enhancement learning is constructed.

[0053] 2. A gait dataset of passengers under natural conditions during the railway station entry process was created, and through model experiments, the performance of the proposed improved gait recognition algorithm is better than the benchmark GaitSet algorithm.

[0054] like Figure 1 As shown in the figure, this application provides an improved gait recognition method, the method including:

[0055] Step S10: Construct a dataset of natural gait of passengers entering railway stations;

[0056] Step S20: Construct an improved gait recognition network model based on GaitSet. The improved gait recognition network model includes a multi-scale feature fusion module and a key feature enhancement module. The multi-scale feature fusion module is used to fuse shallow detail feature information and deep semantic feature information of the convolutional neural network. The key feature enhancement module is used to enhance the key features of gait information.

[0057] Gait recognition model training step S30: Using the passenger natural gait dataset, CASIA-B gait dataset and GREW gait dataset, train the improved gait recognition network model to verify the recognition rate of the improved gait recognition network model.

[0058] In a specific embodiment of the present invention, step S20 of constructing the gait recognition network based on GaitSet further includes:

[0059] Feature acquisition steps: Input the gait sequence into the first convolutional neural network and pooling layer of the benchmark model to obtain the first layer of features; input the first layer of features into the second convolutional neural network and pooling layer of the benchmark model to obtain the second layer of features;

[0060] Multi-scale feature fusion steps: Input the first layer features into the multi-scale feature fusion module, perform downsampling to obtain the first sampled value, fuse it with the second layer features to obtain the first fused feature, which is used as the input and relearning of the subsequent neural network.

[0061] In a specific embodiment of the present invention, step S20 of constructing a gait recognition network based on GaitSet further includes:

[0062] The re-fusion step is as follows: The first fused features are input into the third layer of the convolutional neural network of the baseline model to obtain the third layer features; the first fused features are input into the multi-scale feature fusion module to perform downsampling to obtain the second sampled values, which are then fused with the third layer features to obtain the second fused features.

[0063] In a specific embodiment of the present invention, step S20 of constructing a gait recognition network based on GaitSet further includes:

[0064] Key feature enhancement steps: After performing pooling on the second fused features to form the fourth layer features, the fourth layer features are input into the key feature enhancement module for feature enhancement; the fourth layer features are fused with the output of MGP and then input into the key feature enhancement module for feature enhancement.

[0065] The key feature enhancement module enhances the feature details of the gait sequence at the pixel level through a spatial attention mechanism, and outputs the enhanced features to perform subsequent operations.

[0066] In this embodiment of the invention, the above-mentioned step S10 of constructing the sample set further includes:

[0067] Collect and create the railway industry gait recognition dataset RENG.

[0068] In this embodiment of the invention, the above-mentioned key feature enhancement step further includes:

[0069] Initialization steps: Perform max pooling and average pooling operations on the input augmented features in the channel dimension to obtain feature maps of size 1×h×w, where 1 represents the number of channels, and h and w are the height and width, respectively.

[0070] Channel stacking step: Stack the 1×h×w feature maps in the channel dimension, input them into a convolutional network for fusion learning, and obtain a feature map of size 1×h×w. Then, perform a Sigmoid activation operation and multiply it with the original feature map to obtain the enhanced feature.

[0071] In this embodiment of the invention, the above-mentioned multi-scale fusion feature f i+1 :

[0072] f i+1 =B(f i )+f i , where f i For the features of the current layer, f i+1 f i The next layer of features, B is a convolutional layer.

[0073] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings:

[0074] 1. Improved network model

[0075] Based on GaitSet, this invention proposes a gait recognition algorithm based on multi-scale feature fusion and key feature enhancement, with the network structure as follows: Figure 2As shown in the diagram, Block{1, 2, 3} represents different layers of the convolutional neural network, and SP stands for Set Pooling, which is the order-independent function mentioned above. Applying the SP operation to the sequence features obtained from the convolutional network yields set-level features. MGP (Multilayer Global Pipeline) and HPM (Horizontal Pyramid Mapping) are also modules in GaitSet; their specific implementations and functions can be found in the original text and existing technical literature. The following section mainly introduces the innovative method of this invention.

[0076] In the GaitSet algorithm, the existing technology proposes the MGP module. Its premise is that different feature layers in a convolutional neural network contain different information; the shallower the layer, the more focused on detailed information, while the deeper the layer, the more focused on overall information. By fusing features from different layers, both detailed features and global information can be considered simultaneously.

[0077] As convolutional neural networks become deeper, the information in the input data diminishes, resulting in fewer target features learned by subsequent layers. To address this issue, this invention proposes a multi-scale feature learning method. For example... Figure 2 As shown, the gait sequence is processed through Block 1 to obtain feature f1. The shallow feature f1 contains detailed information about the target. The feature obtained by passing f1 through Block 2 and the feature after downsampling f1 are added together to obtain f2. Finally, feature f3 is obtained using the same method. It can be expressed as follows:

[0078] f i+1 =B(f i )+f i (1)

[0079] Where f i+1 f i The next level of features, B represents the convolutional layer. Furthermore, the addition operation in the formula can be replaced with other operations, such as feature channel stacking.

[0080] The method of this invention can reduce the decay rate of features and enable the network to learn information at different scales, thereby increasing the network's feature extraction capability.

[0081] Unlike person re-identification and image classification, gait data lacks visual features. While this avoids the influence of similar visual features in person re-identification, the limited gait information still presents a significant challenge for recognition. Especially with variations in clothing, gait data for the same person can differ greatly, while different people may have very similar gaits. In such cases, accurately distinguishing them using common methods becomes extremely difficult. This invention recognizes that capturing and enhancing the detailed information of the gait in these situations can effectively alleviate the aforementioned problems. Based on the above analysis, this invention proposes a Feature Enhancement Module (FEM) to perform enhancement learning on the obtained features, and then feeds the enhanced features into subsequent modules. The detailed representation of the FEM module is as follows: Figure 3 As shown.

[0082] The FEM module mainly consists of a spatial attention mechanism, which focuses on gait details. It enhances these details at the pixel level to achieve fine-grained recognition. This process can be represented by the following formula:

[0083] f o =SA(f i )×f i (2)

[0084] Where f i This represents the input features, f o For the enhanced features, SA represents the spatial attention mechanism, where the spatial attention mechanism is as follows: Figure 4 As shown.

[0085] First, the input features are subjected to max pooling and average pooling operations along the channel dimension, resulting in feature maps of size 1×h×w, where 1 represents the number of channels, and h and w are the height and width, respectively, identical to the height and width of the input features. Then, these two 1×h×w feature maps are superimposed along the channel dimension and fed into a convolutional network for fusion learning, yielding a new 1×h×w feature map. Finally, this map undergoes a sigmoid activation operation and is multiplied by the original feature map to obtain the enhanced output feature. The above process can be expressed by the following formula:

[0086] f o =f i *Sig(Conv(Concat(MP(f i ), AP(f i )))) ⑶

[0087] In the above formula, f i and f oThese represent the input features and the processed output features, respectively. MP and AP represent max pooling and average pooling operations in the spatial dimension, respectively. Concat represents the stacking operation in the channel dimension. Conv represents the convolution operation. Sig represents the sigmoid activation function.

[0088] 2. RENG Railway Station Passenger Gait Dataset

[0089] Most current gait recognition research is based on datasets collected under laboratory conditions, such as CASIA-B and OU-MVLP. Researchers have found that gait data collected in these environments is limited, failing to account for various interferences in real-world application environments, which is detrimental to practical applications. Zhu Zheng et al. from Tsinghua University proposed the first open-environment gait dataset, GREW. Experimental results show that most algorithms that perform well under laboratory conditions show poor accuracy on this dataset. To promote the practical application of gait recognition algorithms and improve their ability to handle open environments, this invention collects and creates the Railway Enter Natural Gait (RPNG) dataset, which depicts gait sequences from 799 individuals, with each individual having more than two sequences. The specific distribution is as follows: Figure 5 As shown.

[0090] As shown in Table 1, 555 individuals (69.4%) had two gait sequences, and 98.5% had six or fewer sequences. The total number of sequences was 2051. Compared to current mainstream gait datasets, RENG's data size is relatively small at present and will continue to expand. Some data are shown below. Figure 5 As shown.

[0091]

[0092] Table 1

[0093] 3. Model training and ablation experiments

[0094] This invention first pre-trains on the gait dataset CASIA-B, and then trains and tests on the proposed gait dataset RPNG. CASIA-B is one of the more mainstream gait datasets, collected and created by the Institute of Automation, Chinese Academy of Sciences. This dataset includes 124 individuals, labeled (001, 002, ..., 124), with each individual having three pedestrian states, each with 11 viewpoints (0°, 18°, ..., 180°). Pedestrian states include normal (6 sequences), backpack (2 sequences), and wearing a coat (2 sequences). Therefore, each individual has 11 × (6 + 2 + 2) = 110 sequences, for a total of 110 × 124 = 13640 sequences.

[0095] During training, the proposed model is first pre-trained on the CASIA-B dataset, and then trained and tested on the constructed RPNG dataset. The ratio of training set to test set data is approximately 2:3, that is, the sequences of 600 people are used as the training set, and the sequences of the remaining 199 people are used as the test set.

[0096] In all experiments, the gait data input to the network had a size of 30×1×64×44, where 30 is the length of the input gait sequence, 1 is the number of channels, 64 is the height of a single gait image, and 44 is the width. During the pre-training phase, Adam was chosen as the optimizer with a learning rate (lr) of 1e-4, and a ternary loss function was used for 80,000 iterations on a 4-card NVIDIA 2080Ti. The batch size was set to (8×16), where 8 represents the number of different pedestrians and 16 is the number of gait sequences for each pedestrian. After pre-training, the pre-trained model parameters were loaded and trained on the dataset of this invention, with the learning rate set to 1e-4 and other parameters kept constant, for another 20,000 iterations.

[0097] During testing, this invention selected sequences from 199 individuals in the RPNG gait dataset as the test set, forming a probe set from any segment of each individual's sequence, and using the remaining sequences as a gallery set.

[0098] The proposed model and the benchmark GaitSet algorithm were subjected to ablation experiments on three datasets: CASIA-B, GREW, and RENG. For the RENG gait dataset, since there are no specific angles or state divisions, a gait test object probe was constructed from any sequence of each person in the test set, and the other sequences were used as the gait base library gallery. During testing, sequences from 199 people in the RENG gait dataset were selected as the test set, with any sequence of each person forming the probe and the other sequences serving as the gallery set.

[0099]

[0100] Table 2 Algorithm Ablation Experiment

[0101] As shown in Table 2, the proposed method achieves better results than the benchmark method on all three datasets. Specifically, on CASIA-B, the proposed method achieves a rank of 95.81%, which is 0.81% higher than the benchmark method Gaitset. On GREW, the proposed method achieves a rank of 48.2%, which is 1.9% higher than the benchmark method. On the self-built RENG dataset, the proposed method achieves a rank of 79.9%, which is 1.51% higher than the benchmark method. The experimental results on these three datasets demonstrate the effectiveness of the proposed method. Furthermore, to demonstrate the effectiveness of each module of the proposed method, ablation experiments were also conducted. When only the multi-scale feature learning module is added or only the feature enhancement module is used, the proposed method (multi-scale feature fusion only) and the proposed method (feature enhancement only) both achieve higher rank results than the benchmark method on the three gait datasets, proving the effectiveness of the improved method.

[0102] Secondly, embodiments of this application provide a gait recognition improvement system, employing the above-mentioned gait recognition improvement method, such as... Figure 6 As shown, the system includes:

[0103] Module 10: Constructing a dataset of passengers' natural gait during the railway station entry process;

[0104] Module 20: Constructing an improved gait recognition network model based on GaitSet. The improved gait recognition network model includes a multi-scale feature fusion module and a key feature enhancement module. The multi-scale feature fusion module is used to fuse shallow detail feature information and deep semantic feature information of the convolutional neural network. The key feature enhancement module is used to enhance the key features of gait information.

[0105] Gait recognition model training module 30: Using the passenger natural gait dataset, CASIA-B gait dataset and GREW gait dataset, train the improved gait recognition network model and verify the recognition rate of the improved gait recognition network model.

[0106] Thirdly, embodiments of this application provide a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the gait recognition improvement method as described above.

[0107] Fourthly, embodiments of this application provide a computer-readable storage medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the gait recognition improvement method as described above.

[0108] In addition, combined Figure 1The improved gait recognition method described in this application can be implemented by a computer device. Figure 7 This is a schematic diagram of the hardware structure of a computer device according to an embodiment of this application.

[0109] The computer device may include a processor 81 and a memory 82 storing computer program instructions.

[0110] Specifically, the processor 81 may include a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of this application.

[0111] The memory 82 may include a mass storage device for data or instructions. For example, and not limitingly, the memory 82 may include a hard disk drive (HDD), a floppy disk drive, a solid-state drive (SSD), flash memory, an optical disk drive, a magneto-optical disk drive, magnetic tape, or a Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, the memory 82 may include removable or non-removable (or fixed) media. Where appropriate, the memory 82 may be internal or external to a data processing device. In a particular embodiment, the memory 82 is non-volatile memory. In a particular embodiment, the memory 82 includes read-only memory (ROM) and random access memory (RAM). Where appropriate, the ROM may be a mask-programmed ROM, a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), an electrically alterable read-only memory (EAROM), or flash memory, or a combination of two or more of these. Where appropriate, the RAM can be Static Random-Access Memory (SRAM) or Dynamic Random-Access Memory (DRAM). DRAM can be Fast Page Mode Dynamic Random-Access Memory (FPMDRAM), Extended Data Out Dynamic Random-Access Memory (EDODRAM), Synchronous Dynamic Random-Access Memory (SDRAM), etc.

[0112] The memory 82 can be used to store or cache various data files that need to be processed and / or communicated, as well as possible computer program instructions executed by the processor 81.

[0113] The processor 81 implements any of the gait recognition improvement methods in the above embodiments by reading and executing computer program instructions stored in the memory 82.

[0114] In some embodiments, the computer device may further include a communication interface 83 and a bus 80. For example, Figure 7 As shown, the processor 81, memory 82, and communication interface 83 are connected through bus 80 and complete communication with each other.

[0115] The communication interface 83 is used to enable communication between the various modules, devices, units, and / or equipment in the embodiments of this application. The communication port 83 can also enable data communication with other components such as external devices, image / data acquisition devices, databases, external storage, and image / data processing workstations.

[0116] Bus 80 includes hardware, software, or both, that couples components of a computer device together. Bus 80 includes, but is not limited to, at least one of the following: data bus, address bus, control bus, expansion bus, and local bus. For example, and not as a limitation, bus 80 may include an Accelerated Graphics Port (AGP) or other graphics bus, an Extended Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), a Hyper Transport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an InfiniBand interconnect, a Low Pin Count (LPC) bus, a memory bus, a Micro Channel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local Bus (VLB) bus, or other suitable buses, or a combination of two or more of these. Where appropriate, bus 80 may include one or more buses. Although specific buses are described and illustrated in the embodiments of this application, this application considers any suitable bus or interconnection.

[0117] Compared with existing technologies, the present invention proposes a new deep learning network model based on the GaitSet gait recognition algorithm, constructs a new gait recognition algorithm based on multi-scale fusion and feature enhancement learning, creates a gait dataset of passengers under natural conditions in the railway station entry process, and demonstrates through model experiments that the performance of the proposed improved gait recognition algorithm is superior to the benchmark GaitSet algorithm.

[0118] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. An improved gait recognition method, characterized in that, The method includes: Steps for constructing the sample set: Construct a dataset of passengers' natural gait during the railway station entry process; The steps for constructing an improved gait recognition network model based on GaitSet are as follows: The improved gait recognition network model includes a multi-scale feature fusion module and a key feature enhancement module. The multi-scale feature fusion module is used to fuse shallow detail feature information and deep semantic feature information from a convolutional neural network; the key feature enhancement module is used to enhance the key features of the gait information. Gait recognition model training steps: Using the aforementioned passenger natural gait dataset, CASIA-B gait dataset, and GREW gait dataset, train the improved gait recognition network model to verify the recognition rate of the improved gait recognition network model; The steps for constructing the gait recognition network based on GaitSet also include: Feature acquisition steps: Input the gait sequence into the first convolutional neural network and pooling layer of the benchmark model to obtain the first layer of features; input the first layer of features into the second convolutional neural network and pooling layer of the benchmark model to obtain the second layer of features; Multi-scale feature fusion step: Input the first layer features into the multi-scale feature fusion module, perform downsampling to obtain the first sampled value, fuse it with the second layer features to obtain the first fused feature, which is used for subsequent neural network input and relearning; Re-fusion step: Input the first fused features into the third layer of the convolutional neural network of the baseline model to obtain the third layer features; input the first fused features into the multi-scale feature fusion module, perform downsampling to obtain the second sampled value, and fuse it with the third layer features to obtain the second fused features; Key feature enhancement steps: After performing pooling operation on the second fused features to form a fourth layer of features, the fourth layer of features is input into the key feature enhancement module for feature enhancement; the features fused with the output of MGP are input into the key feature enhancement module for feature enhancement; the key feature enhancement module enhances the feature details of the gait sequence at the pixel level through a spatial attention mechanism, and outputs the enhanced features for subsequent operations.

2. The improved gait recognition method according to claim 1, characterized in that, The step of constructing the sample set further includes: Collect and create the railway industry gait recognition dataset RENG.

3. The improved gait recognition method according to claim 1, characterized in that, The key feature enhancement step further includes: Initialization steps: Perform max pooling and average pooling operations on the input enhanced features respectively to obtain feature maps of size 1×h×w, where 1 represents the number of channels, and h and w are the height and width, respectively; Channel overlay step: The 1×h×w feature maps are overlaid along the channel dimension, input into a convolutional network for fusion learning, and a feature map of size 1×h×w is obtained. The activation operation is performed and multiplied with the original feature map to obtain the enhanced feature.

4. The improved gait recognition method according to claim 3, characterized in that, Multi-scale feature fusion ,in, for The next layer of features, It is a convolutional layer.

5. An improved gait recognition system, employing the improved gait recognition method as described in any one of claims 1-4, characterized in that, The system includes: The sample set construction module is used to construct a dataset of passengers' natural gait during the railway station entry process. A module for constructing an improved gait recognition network model based on GaitSet is included. This improved gait recognition network model comprises a multi-scale feature fusion module and a key feature enhancement module. The multi-scale feature fusion module is used to fuse shallow detail feature information and deep semantic feature information from a convolutional neural network. The key feature enhancement module is used to enhance the key features of gait information. Gait recognition model training module: Using the aforementioned passenger natural gait dataset, CASIA-B gait dataset, and GREW gait dataset, the improved gait recognition network model is trained to verify the recognition rate of the improved gait recognition network model; The construction of the gait recognition network module based on GaitSet improvement also includes: Feature acquisition module: Input the gait sequence into the first convolutional neural network and pooling layer of the benchmark model to obtain the first layer of features; input the first layer of features into the second convolutional neural network and pooling layer of the benchmark model to obtain the second layer of features; Multi-scale feature fusion module: The first layer features are input into the multi-scale feature fusion module, downsampling is performed to obtain the first sampled value, which is then fused with the second layer features to obtain the first fused feature, which is used as the input and relearning of the subsequent neural network; The re-fusion module: inputs the first fused features into the third layer of the convolutional neural network of the baseline model to obtain the third layer features; inputs the first fused features into the multi-scale feature fusion module, performs downsampling to obtain the second sampled value, and fuses it with the third layer features to obtain the second fused features; Key Feature Enhancement Module: After performing pooling operation on the second fused features to form a fourth layer of features, the fourth layer of features is input into the key feature enhancement module for feature enhancement; the features fused with the fourth layer of features and the output of MGP are input into the key feature enhancement module for feature enhancement; the key feature enhancement module enhances the feature details of the gait sequence at the pixel level through a spatial attention mechanism, and outputs the enhanced features for subsequent operations.

6. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the improved gait recognition method as described in any one of claims 1 to 4.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the improved gait recognition method as described in any one of claims 1 to 4.

Images