A method, device, equipment and medium for generating a gait recognition algorithm deployment strategy

Abstract

The present application provides a method, device, equipment and medium for generating a gait recognition algorithm deployment strategy. Among them, the method includes: the trained target decision tree model outputs a decision tree image according to a plurality of parameter combinations used during its training; the parameter combinations include at least one of acquisition device deployment parameters, shooting environment parameters, video image attribute parameters, and shooting system parameters; extract the influence degree of each parameter in the parameter combination on the gait recognition similarity from the decision tree image, and select the parameters ranked in the top target number as target parameters in the order of the influence degree of each parameter from large to small; generate a deployment strategy according to the target parameters for shooting according to the deployment strategy when establishing a target gait feature database. Through this method, it is beneficial to improve the accuracy of gait recognition.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular, to a method, device, equipment and medium for generating a deployment strategy for a gait recognition algorithm. Background Art

[0002] Biometric recognition technology is a method for identifying an individual's identity. It uses the inherent physiological or behavioral characteristics of the human body for individual identity authentication. Biometric recognition technology includes various recognition technologies such as face recognition, fingerprint recognition, iris recognition, and gait recognition. Since each person's biometric characteristics are unique and universal, and are not easily forged or counterfeited, using biometric recognition technology for identity authentication has the advantages of being safe, reliable, and accurate.

[0003] Gait recognition aims to identify a person's identity based on their walking posture. As the second-generation biometric recognition technology, gait recognition is the only biometric recognition technology that can perform identity authentication at a long distance. It has the advantages of good concealment, low requirements for video quality, long-distance non-contact, and difficulty in being disguised. Based on the above advantages, gait recognition has attracted much attention in recent years and has broad application prospects in the field of visual surveillance.

[0004] In the prior art, a gait recognition algorithm is usually used to compare the gait characteristics of a target person collected with the gait characteristics already existing in a gait characteristic database, and calculate the similarity between the two. The level of similarity is the basis for the credibility of identity recognition. However, in the actual application process, the accuracy of gait recognition is not only affected by the gait recognition algorithm itself, but also affected by factors such as image acquisition equipment, on-site environment, and the quality of captured gait information when establishing the gait characteristic database, thereby affecting the accuracy of gait recognition. Summary of the Invention

[0005] In view of this, the purpose of this application is to provide a method, device, equipment and medium for generating a deployment strategy for a gait recognition algorithm, which analyzes environmental parameters, device parameters, and captured gait information parameters that affect the accuracy of gait recognition, so as to improve the accuracy of the gait recognition algorithm at the application site.

[0006] In a first aspect, an embodiment of this application provides a method for generating a deployment strategy for a gait recognition algorithm, including:

[0007] A trained target decision tree model outputs a decision tree image according to multiple parameter combinations used in the training process; the parameter combinations include at least one of acquisition device deployment parameters, shooting environment parameters, video image attribute parameters, and shooting system parameters;

[0008] Extract the influence degree of each parameter in the parameter combination on the gait recognition similarity from the decision tree image, and select the parameters located in the top target number as target parameters in the order of the influence degree of each parameter from large to small;

[0009] Generate a deployment strategy according to the target parameters, so as to perform shooting according to the deployment strategy when establishing the target gait feature database.

[0010] Combined with the first aspect, the embodiment of the present application provides a first possible implementation manner of the first aspect, wherein the target decision tree model is trained in the following manner:

[0011] Obtain a plurality of training sample data; the training sample data includes the parameter combination and the first similarity corresponding to the parameter combination; the first similarity is obtained according to the comparison similarity between the captured gait feature of the target object and the sample gait feature. When the comparison similarity is greater than the first preset similarity, the first similarity corresponding to the comparison similarity is 1. When the comparison similarity is not greater than the first preset similarity, the first similarity corresponding to the comparison similarity is 0; the sample gait feature is extracted from the first sample image obtained when shooting using the first deployment strategy; the first deployment strategy is generated according to the parameter combination; the captured gait feature is extracted from the second sample image obtained when randomly shooting the target object;

[0012] Divide the training sample data into a training set and a validation set;

[0013] Input the parameter combination in the training set into at least one initial decision tree model to be trained, and each initial decision tree model outputs the second similarity corresponding to the parameter combination; the value of the second similarity is 0 or 1;

[0014] For each initial decision tree model, calculate the loss function using the first similarity and the second similarity output by the initial decision tree model to obtain a loss value, and use the loss value to perform backpropagation training on the initial decision tree model to adjust the learnable parameters in the initial decision tree model until all the training sample data in the training set are used up and stop training, and use the current initial decision tree model as the intermediate decision tree model;

[0015] Input the parameter combination in the validation set into each intermediate decision tree model, and each intermediate decision tree model outputs the third similarity corresponding to the parameter combination; the value of the third similarity is 0 or 1;

[0016] For each of the intermediate decision tree models, calculate the accuracy of the intermediate decision tree model according to the first similarity and the third similarity output by the intermediate decision tree model;

[0017] According to the accuracy of each of the intermediate decision tree models, select the intermediate decision tree model with the highest accuracy from all the intermediate decision tree models as the target decision tree model.

[0018] Combined with the first aspect, the embodiments of the present application provide a second possible implementation manner of the first aspect, wherein generating a deployment strategy according to the target parameter and performing shooting according to the deployment strategy when establishing a target gait feature database includes:

[0019] Obtain the first gait feature and the second gait feature of the test object; the first gait feature is extracted from the first test image obtained by shooting the test object using the deployment strategy; the second gait feature is extracted from the second test image obtained by randomly shooting the test object;

[0020] Calculate the fourth similarity between the first gait feature and the second gait feature, and when the fourth similarity is greater than the second preset similarity, perform shooting according to the deployment strategy when establishing a target gait feature database.

[0021] Combined with the first aspect, the embodiments of the present application provide a third possible implementation manner of the first aspect, wherein the deployment strategy includes a target shooting duration when shooting a specified object; the method further includes:

[0022] During the process of shooting according to the deployment strategy for the purpose of establishing a target gait feature database, when the shooting duration when shooting the specified object is less than the target shooting duration, send a prompt message to the specified object and reshoot the specified object; the prompt message is used to prompt the specified object to slow down its walking speed.

[0023] Combined with the first aspect, the embodiments of the present application provide a fourth possible implementation manner of the first aspect, wherein the acquisition device deployment parameters include at least one of the number and pitch angle of the acquisition devices; the shooting environment parameters include at least one of the complexity and light intensity of the shooting environment; the video image attribute parameters include at least one of the acquisition duration, shooting angle, sequence length of the captured video, and the image clarity, image resolution, and image pixels of each frame image in the captured video; the shooting system parameters include a frame skipping strategy for selecting images from the video.

[0024] In a second aspect, the embodiments of the present application further provide a device for generating a deployment strategy for a gait recognition algorithm, including:

[0025] An output module, configured to output a decision tree image according to a plurality of parameter combinations used during the training of a trained target decision tree model; the parameter combinations include at least one of acquisition device deployment parameters, shooting environment parameters, video image attribute parameters, and shooting system parameters;

[0026] An extraction module, configured to extract the influence degree of each parameter in the parameter combination on the gait recognition similarity from the decision tree image, and select the top target number of parameters as target parameters in the order of the influence degree of each parameter from large to small;

[0027] A generation module, configured to generate a deployment strategy according to the target parameters, so as to perform shooting according to the deployment strategy when establishing a target gait feature database.

[0028] Combined with the second aspect, the embodiments of the present application provide a first possible implementation manner of the second aspect, where, further comprising:

[0029] An acquisition module, configured to acquire a plurality of training sample data; the training sample data includes the parameter combination and a first similarity corresponding to the parameter combination; the first similarity is obtained according to the comparison similarity between the captured gait feature of the target object and the sample gait feature. When the comparison similarity is greater than a first preset similarity, the first similarity corresponding to the comparison similarity is 1. When the comparison similarity is not greater than the first preset similarity, the first similarity corresponding to the comparison similarity is 0; the sample gait feature is extracted from a first sample image obtained when shooting using a first deployment strategy; the first deployment strategy is generated according to the parameter combination; the captured gait feature is extracted from a second sample image obtained when randomly shooting the target object;

[0030] A division module, configured to divide the training sample data into a training set and a validation set;

[0031] A first input module, configured to input the parameter combination in the training set into at least one initial decision tree model to be trained, and each initial decision tree model respectively outputs a second similarity corresponding to the parameter combination; the value of the second similarity is 0 or 1;

[0032] The first calculation module is used to calculate a loss value for each of the initial decision tree models by using the first similarity and the second similarity output by the initial decision tree model, and perform backpropagation training on the initial decision tree model by using the loss value to adjust the learnable parameters in the initial decision tree model until the training stops when all the training sample data in the training set are used up, and use the current initial decision tree model as the intermediate decision tree model;

[0033] The second input module is used to input the parameter combinations in the verification set into each of the intermediate decision tree models, and each of the intermediate decision tree models respectively outputs a third similarity corresponding to the parameter combination; the value of the third similarity is 0 or 1;

[0034] The second calculation module is used to calculate the accuracy of each of the intermediate decision tree models according to the first similarity and the third similarity output by the intermediate decision tree model;

[0035] The selection module is used to select the intermediate decision tree model with the highest accuracy as the target decision tree model from all the intermediate decision tree models according to the accuracy of each of the intermediate decision tree models.

[0036] Combined with the second aspect, the embodiment of the present application provides a second possible implementation manner of the second aspect. When the generation module is used to generate a deployment strategy according to the target parameters and perform shooting according to the deployment strategy when establishing the target gait feature database, it is specifically used for:

[0037] Obtain the first gait feature and the second gait feature of the test object; the first gait feature is extracted from the first test image obtained by shooting the test object using the deployment strategy; the second gait feature is extracted from the second test image obtained by randomly shooting the test object;

[0038] Calculate a fourth similarity between the first gait feature and the second gait feature, and when the fourth similarity is greater than the second preset similarity, perform shooting according to the deployment strategy when establishing the target gait feature database.

[0039] Combined with the second aspect, the embodiment of the present application provides a third possible implementation manner of the second aspect. The deployment strategy includes a target shooting duration when shooting a specified object; the device further includes:

[0040] A prompting module, which is used to send a prompting message to the specified object and reshoot the specified object when the shooting duration is less than the target shooting duration during the shooting according to the deployment strategy for the purpose of establishing a target gait feature database; the prompting message is used to prompt the specified object to slow down its walking speed.

[0041] Combined with the second aspect, the embodiments of the present application provide a fourth possible implementation manner of the second aspect, wherein the acquisition device deployment parameters include at least one of the number of acquisition devices and the pitch angle; the shooting environment parameters include at least one of the complexity of the shooting environment and the light intensity; the video image attribute parameters include at least one of the acquisition duration of the captured video, the shooting perspective, the sequence length, and the image clarity, image resolution, and image pixels of each frame image in the captured video; the shooting system parameters include the frame skipping strategy for selecting images from the video.

[0042] In a third aspect, the embodiments of the present application further provide an electronic device, including: a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the electronic device runs, the processor communicates with the memory through the bus. When the machine-readable instructions are executed by the processor, the steps in any of the possible implementation manners in the first aspect are executed.

[0043] In a fourth aspect, the embodiments of the present application further provide a computer-readable storage medium, on which a computer program is stored. When the computer program is run by a processor, the steps in any of the possible implementation manners in the first aspect are executed.

[0044] A method, device, equipment, and medium for generating a deployment strategy for a gait recognition algorithm provided by the embodiments of the present application determine the influence degree of each parameter in the parameter combination on the gait recognition similarity through a decision tree model, and select the parameters in the top target number as target parameters in the order from large to small according to the influence degree of each parameter, and generate a deployment strategy according to the target parameters for shooting according to the deployment strategy when establishing a target gait feature database. Through this method, before establishing a target gait feature database, the deployment strategy for establishing the target gait feature database is first determined, and shooting is performed according to the deployment strategy when establishing the target gait feature database, which can make the gait features in the established target gait feature database more accurate, thereby improving the accuracy of the gait recognition algorithm at the application site and further improving the accuracy of gait recognition.

[0045] To make the above objects, features, and advantages of the present application more obvious and understandable, the following specific preferred embodiments are given and described in detail in conjunction with the accompanying drawings. Description of the Drawings

[0046] To more clearly illustrate the technical solutions of the embodiments of the present application, the accompanying drawings required for the embodiments will be briefly introduced below. It should be understood that the following accompanying drawings only show some embodiments of the present application and should not be regarded as limiting the scope. For those of ordinary skill in the art, other relevant accompanying drawings can be obtained based on these drawings without creative efforts.

[0047] Figure 1 It shows a flowchart of a method for generating a gait recognition algorithm deployment strategy provided by an embodiment of the present application;

[0048] Figure 2 It shows a schematic diagram of the first similarity generation process provided by an embodiment of the present application;

[0049] Figure 3 It shows a schematic structural diagram of a device for generating a gait recognition algorithm deployment strategy provided by an embodiment of the present application;

[0050] Figure 4 It shows a schematic structural diagram of an electronic device provided by an embodiment of the present application. Detailed implementation manners

[0051] To make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, rather than all the embodiments. Usually, the components of the embodiments of the present application described and shown in the accompanying drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the accompanying drawings is not intended to limit the scope of the present application claimed, but only represents the selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present application.

[0052] For ease of understanding, the application scenario of the present application will be briefly described first. When performing gait recognition on a target person, it is necessary to capture a video of the target person through a camera device, and then extract the gait features of the target person from each consecutive frame image in the video through a gait recognition algorithm, and compare the gait features of the target person with the existing gait features in a pre-established gait feature database, calculate the similarity between the gait features of the target person and the existing gait features in the gait feature database, so as to identify the target person from the gait feature database. In a specific application scenario, when the gait features of the target person are stored in the gait feature database, it means that the target person is an employee of this unit and is allowed to enter this unit.

[0053] Among them, when establishing the gait feature database of the employees of the unit in advance, each employee needs to walk in a specified venue, and the walking video of the employee is recorded simultaneously by one or more acquisition devices. The gait features of the employee are extracted from the consecutive frame images in the video, and the gait features and identity information of the employee are stored in the gait feature database.

[0054] Considering the problem that the accuracy of gait recognition is affected by factors such as the pre-established gait feature database, the image acquisition device and the environment when capturing gait features, which in turn affects the accuracy of the gait recognition algorithm. Based on this, the embodiments of the present application provide a method, device, equipment and medium for generating a gait recognition algorithm deployment strategy. Before establishing the target gait feature database, the deployment strategy when establishing the target gait feature database is determined first, so as to perform shooting according to the deployment strategy when establishing the target gait feature database, thereby improving the accuracy of gait recognition. The following is described through embodiments.

[0055] Embodiment 1:

[0056] For the convenience of understanding this embodiment, first, a method for generating a gait recognition algorithm deployment strategy disclosed in the embodiments of the present application is introduced in detail. Figure 1 The flowchart of a method for generating a gait recognition algorithm deployment strategy provided by the embodiments of the present application is shown. As Figure 1 shown, it includes the following steps S101-S103:

[0057] S101: The trained target decision tree model outputs a decision tree image according to multiple parameter combinations used in the training process; the parameter combinations include at least one of acquisition device deployment parameters, shooting environment parameters, video image attribute parameters, and shooting system parameters.

[0058] In this embodiment, before establishing the final target gait feature database, the sample gait features of the target object under different parameter combinations are established according to different parameter combinations. The parameter combination refers to the combination of shooting parameters at the shooting site when establishing the gait features of the target object, and the parameter combination can be preset according to prior experience. For the same target object, the number of parameter combinations is the same as the number of sample gait features of the target object, and each parameter combination corresponds to a sample gait feature of the target object. The sample gait features corresponding to different parameter combinations may be different.

[0059] Exemplarily, the parameter combinations can be: using several acquisition devices to record the walking video of the target object, and pre-determining what the pitch angle of each acquisition device is, as well as pre-determining the complexity of the background of the shooting scene, etc. At least one parameter is different in each parameter combination. For example, three acquisition devices are used in parameter combination A, and one acquisition device is used in parameter combination B. Among them, the acquisition device can be a camera.

[0060] S102: Extract the influence degree of each parameter in the parameter combination on the gait recognition similarity from the decision tree image, and select the parameters of the top target number as target parameters in the order of the influence degree of each parameter from large to small.

[0061] In this application, this method is applied to the server, and the target decision tree model is set in the server. The target decision tree model outputs a decision tree image after training. After the server obtains the decision tree image, step S102 is executed. The decision tree image is a visual image, and the decision tree image contains the influence degree of each parameter in the parameter combination on the gait recognition similarity. It can be understood that, for example, the decision tree image contains what the influence degree on the gait recognition similarity is when 1 acquisition device is deployed, what the influence degree on the gait recognition similarity is when 3 acquisition devices are deployed, what the influence degree on the gait recognition similarity is when the pitch angle is 30 degrees, etc. Among them, the decision tree image can be an image in PNG format.

[0062] In a specific embodiment, extract the influence degree of each parameter in the parameter combination on the gait recognition similarity and the arrangement order of the influence degree of each parameter on the gait recognition similarity from large to small from the decision tree image, and select the parameters of the top target number as target parameters. Exemplarily, if the sequence length (i.e., a parameter in the parameter combination) is 40 frames and its influence degree on the gait recognition similarity ranks among the top, then the sequence length of 40 frames is used as the target parameter. Or, for example, when the pitch angle of the acquisition device (i.e., another parameter in the parameter combination) is 20 degrees and its influence degree on the gait recognition similarity ranks among the top, then the pitch angle of the acquisition device of 20 degrees is used as the target parameter.

[0063] S103: Generate a deployment strategy according to the target parameters for shooting according to the deployment strategy when establishing the target gait feature database.

[0064] The deployment strategy can include the deployment quantity of the acquisition device, the deployment pitch angle of the acquisition device, the complexity of the shooting environment, the light intensity of the shooting environment, the acquisition duration of the shooting video, etc. Among them, the acquisition device can be a camera. Exemplarily, continuing with the example in step S102, the deployment strategy can be that the sequence length is 40 frames and the pitch angle of the acquisition device is 20 degrees.

[0065] In a possible implementation, the acquisition device deployment parameters include at least one of the number of acquisition devices and the pitch angle; the shooting environment parameters include at least one of the complexity of the shooting environment and the light intensity; the video image attribute parameters include at least one of the acquisition duration of the captured video, the shooting angle, the sequence length, and the image clarity, image resolution, and image pixels of each frame of the captured video; the shooting system parameters include the frame skipping strategy for selecting images from the video.

[0066] Among them, the acquisition device refers to the device used to acquire the video of the target object while walking. The acquisition device deployment parameters refer to the parameters for deploying the acquisition device in the video acquisition scenario. For example, how many acquisition devices are deployed in the video acquisition scenario and what is the pitch angle of each acquisition device. Among them, the pitch angle is the angle between the acquisition device and the horizontal direction.

[0067] The shooting environment refers to the shooting scenario where the target object is located when the acquisition device acquires the walking video of the target object. The shooting environment parameters refer to the parameters of the shooting scenario where the target object is located when the acquisition device acquires the walking video of the target object. Among them, the complexity of the shooting environment can be the color of the ground in the shooting environment, the color of the target object's clothes, etc. The light intensity of the shooting environment can be the intensity of the light in the shooting scenario when acquiring the walking video of the target object, such as daytime, nighttime, sunny, cloudy, morning, afternoon, etc.

[0068] The captured video refers to the walking video of the target object in the shooting scenario captured by the acquisition device. The acquisition duration of the captured video refers to the time used during the process of the target object walking from one specified position to another specified position when acquiring the walking video of the target object. That is to say, the walking route of the target object is fixed. Since the walking speeds of each target object are different, the acquisition duration of each target object is different. The shooting angle refers to the angle between the target object and the acquisition device. For example, is the shooting angle between the target object and the acquisition device a face-to-face shooting angle or a shooting angle with the back of the target object facing the acquisition device. The sequence length refers to the number of frame images in the video stream of the captured video acquired by the acquisition device during the process of the target object walking from one specified position to another specified position. When the target object walks faster, the sequence length is less, such as 10 frames. When the target object walks slower, the sequence length is more, such as 40 frames. Each frame image in the video refers to each frame image in the video stream of the captured video.

[0069] The acquisition device transmits the captured video to the shooting system. The frame skipping strategy refers to the strategy of how many frames (i.e., how many frames to skip) are selected from all the frame images in the video stream of the received shooting video by the shooting system. For example, one image is selected by skipping 2 frames from all the frame images.

[0070] In a possible implementation manner, the target decision tree model is trained in the following way:

[0071] S1001: Obtain a plurality of training sample data; the training sample data includes a parameter combination and a first similarity corresponding to the parameter combination; the first similarity is obtained according to the comparison similarity between the captured gait feature of the target object and the sample gait feature. When the comparison similarity is greater than the first preset similarity, the first similarity corresponding to the comparison similarity is 1. When the comparison similarity is not greater than the first preset similarity, the first similarity corresponding to the comparison similarity is 0; the sample gait feature is extracted from the first sample images obtained when shooting using the first deployment strategy; the first deployment strategy is generated according to the parameter combination; the captured gait feature is extracted from the second sample images obtained when randomly shooting the target object.

[0072] In this embodiment, Figure 2 shows a schematic diagram of the first similarity generation process provided by the embodiment of the present application, as Figure 2 shown. For each parameter combination, the first deployment strategy of the parameter combination is generated according to the parameter combination. According to the first deployment strategy, a scenario for establishing a gait feature database is built. For example, the first deployment strategy is that the number of acquisition devices is 2, the pitch angles are 30 degrees and 60 degrees respectively, and the sequence length is 20 pictures, etc. Organize a certain amount of target objects to walk on a specified route, and shoot the walking target objects through the acquisition device to obtain the first sample video of the target object when walking; according to the frame skipping strategy in the first deployment strategy, multiple first sample images are selected from the first sample video, and the sample gait feature of the target object is extracted from the multiple first sample images. Therefore, for the same target object, each parameter combination corresponds to a sample gait feature of the target object, and the number of parameter combinations is the same as the number of sample gait features of the target object. Since there are multiple parameter combinations, there are multiple sample gait features of the target object.

[0073] When the target object walks in any other environment, randomly capture the second sample video of the target object when walking, select multiple second sample images from the second sample video according to the frame skipping strategy in the first deployment strategy, and extract the captured gait feature of the target object from the multiple second sample images. In this embodiment, there is only one captured gait feature of the target object.

[0074] For each gait sample feature of the target object, input the gait sample feature and the captured gait feature of the target object into a pre-trained gait recognition algorithm, and output the comparison similarity between the gait sample feature and the captured gait feature of the target object. The value range of the comparison similarity is from 0 to 1, and the larger the value, the higher the similarity, that is, the more similar. Compare the comparison similarity with a first preset similarity. When the comparison similarity is greater than the first preset similarity, the first similarity corresponding to the comparison similarity is 1. When the comparison similarity is not greater than the first preset similarity, the first similarity corresponding to the comparison similarity is 0. Each gait sample feature corresponds to a first similarity, that is, each parameter combination corresponds to a first similarity. Since there are multiple parameter combinations, each target object corresponds to multiple first similarities.

[0075] S1002: Divide the training sample data into a training set and a validation set.

[0076] In this embodiment, the training sample data can be divided into a training set and a validation set according to a preset ratio.

[0077] Exemplarily, when the number of training sample data is 100 and the preset ratio is 50%, the training set includes 50 training sample data, and the validation set includes 50 training sample data. Each training sample data contains a parameter combination and the first similarity corresponding to the parameter combination.

[0078] S1003: Input the parameter combinations in the training set into at least one initial decision tree model to be trained, and each initial decision tree model respectively outputs the second similarity corresponding to the parameter combination; the value of the second similarity is 0 or 1.

[0079] Pre-select at least one initial decision tree model to be trained, such as including C4.5 decision tree, CART decision tree (classification and regression tree), or ID3 decision tree, etc.

[0080] In the embodiment of the present application, each parameter combination in the training set is input into at least one initial decision tree model to be trained, and each initial decision tree model respectively outputs the second similarity corresponding to the parameter combination predicted by the initial decision model.

[0081] S1004: For each initial decision tree model, calculate the loss function using the first similarity and the second similarity output by the initial decision tree model to obtain a loss value, and use the loss value to perform backpropagation training on the initial decision tree model to adjust the learnable parameters in the initial decision tree model until the training stops when all the training sample data in the training set have been used, and use the current initial decision tree model as an intermediate decision tree model.

[0082] Use the first similarity corresponding to the parameter combination as a label to train each initial decision tree model. When the second similarity corresponding to the parameter combination is the same as the first similarity corresponding to the parameter combination, it indicates that the initial decision tree model makes a correct prediction.

[0083] For each initial decision tree model, after all the training sample data in the training set have been used up, that is, after each training sample data in the training set has completed the training of the initial decision tree model, use the current initial decision tree model as an intermediate decision tree model.

[0084] S1005: Input the parameter combinations in the validation set into each intermediate decision tree model, and each intermediate decision tree model respectively outputs the third similarity corresponding to the parameter combination; the value of the third similarity is 0 or 1.

[0085] After using the training sample data in the training set to train the initial decision tree model to obtain an intermediate decision tree model, it is necessary to verify the intermediate decision tree model and calculate the accuracy of the intermediate decision tree model.

[0086] S1006: For each intermediate decision tree model, calculate the accuracy of the intermediate decision tree model according to the first similarity and the third similarity output by the intermediate decision tree model.

[0087] Exemplarily, when the first similarity is the same as the third similarity output by the intermediate decision tree model, it indicates that the intermediate decision tree model makes a correct prediction; when the first similarity is different from the third similarity output by the intermediate decision tree model, it indicates that the intermediate decision tree model makes a wrong prediction. Specifically, calculate the ratio of the number of correct predictions to the total number of predictions (the total number of predictions includes the number of correct predictions and the number of wrong predictions), and use this ratio as the accuracy of the intermediate decision tree model.

[0088] S1007: Select the intermediate decision tree model with the highest accuracy from all the intermediate decision tree models according to the accuracy of each intermediate decision tree model, and use the intermediate decision tree model with the highest accuracy as the target decision tree model.

[0089] Select the intermediate decision tree model with the highest accuracy from all the intermediate decision tree models according to the accuracy of each intermediate decision tree model, so as to use the intermediate decision tree model with the highest accuracy as the target decision tree model.

[0090] In a possible implementation manner, when performing step S103 to generate a deployment policy according to the target parameters and perform shooting according to the deployment policy when establishing the target gait feature database, it may be specifically performed according to the following steps:

[0091] S1031: Obtain the first gait feature and the second gait feature of the test object; the first gait feature is extracted from the first test image obtained by photographing the test object using the deployment strategy; the second gait feature is extracted from the second test image obtained when randomly photographing the test object.

[0092] After generating the deployment strategy according to the target parameters, in order to ensure that using this deployment strategy for photographing when establishing the target gait feature database can improve the accuracy of gait recognition, in this embodiment, the test object is first photographed according to this deployment strategy, aiming to test whether the gait data features of the test object obtained when using this deployment strategy to photograph the test object are standard enough.

[0093] Specifically, adjust the acquisition device and the shooting environment according to the data such as the number of acquisition devices, pitch angle, etc. included in the deployment strategy, and the shooting environment parameters. Then, in this shooting environment, record the first test video of the test object walking through the adjusted acquisition device. According to the frame skipping strategy included in the deployment strategy, select multiple first test images from the first test video, and extract the first gait feature of the test object from the multiple first test images. In any scenario, record the second test video of the test object walking through the arbitrarily deployed acquisition device. According to the frame skipping strategy included in the deployment strategy, select multiple second test images from the second test video, and extract the second gait feature of the test object from the multiple second test images.

[0094] S1032: Calculate the fourth similarity between the first gait feature and the second gait feature. When the fourth similarity is greater than the second preset similarity, photograph according to the deployment strategy when establishing the target gait feature database.

[0095] Input the first gait feature and the second gait feature into the pre-trained gait recognition algorithm, and calculate the fourth similarity between the first gait feature and the second gait feature through the gait recognition algorithm. The value range of the fourth similarity is from 0 to 1. When the fourth similarity is greater than the second preset similarity, it means that when using this deployment strategy to photograph the test object, the gait data features of the test object obtained are standard enough. At this time, the target gait features in the target gait feature database obtained by photographing according to the deployment strategy when establishing the target gait feature database are standard enough.

[0096] In a specific embodiment, the first gait feature and the second gait feature of multiple test objects can be obtained. After obtaining the fourth similarity of each test object through the above process, it is determined whether the fourth similarity of each test object is greater than the second preset similarity. When the fourth similarity of each test object is greater than the second preset similarity, it indicates that when using this deployment strategy to photograph the test object, the gait data features of the test object obtained are standard enough.

[0097] In this embodiment, the specific process of photographing according to the deployment strategy when establishing the target gait feature database can be as follows: According to parameters such as the number of acquisition devices, pitch angle, and shooting environment included in the deployment strategy, deploy the acquisition devices and the shooting scene, and then let the designated object walk along the designated route in the shooting scene. The acquisition device records the target video of the designated object during the walking process, selects multiple target gait images from the target video according to the frame skipping strategy included in the deployment strategy, and extracts the target gait features of the designated object from the multiple target gait images. In this embodiment, each designated object corresponds to a target gait feature. Store the target gait features of all designated objects in the target gait feature database. In a specific embodiment, the designated object can be an employee of a company, etc.

[0098] In another possible case, calculate the fourth similarity between the first gait feature and the second gait feature. When the fourth similarity is not greater than the second preset similarity, re - execute steps S101 - S103 to regenerate the deployment strategy.

[0099] In a possible implementation manner, the deployment strategy includes the target shooting duration when photographing a designated object; the method further includes:

[0100] During the process of photographing according to the deployment strategy for the purpose of establishing the target gait feature database, when the shooting duration when photographing the designated object is less than the target shooting duration, send a prompt message to the designated object and re - photograph the designated object; the prompt message is used to prompt the designated object to slow down its walking speed.

[0101] When the shooting duration when photographing the designated object is less than the target shooting duration, it indicates that the walking speed of the designated object is too fast. Therefore, by sending a prompt message to the designated object, to prompt the target object to slow down its walking speed, and re - photograph the process of the designated object walking on the designated route.

[0102] In the embodiment of the present application, another application scenario is also provided. After constructing the target gait feature database, when using the target gait feature database to perform gait recognition on a designated object, it can be specifically executed in the following manner:

[0103] Obtain the first training sample data; the first training sample data includes the first captured image parameters and the fifth similarity corresponding to the first captured image parameters; the fifth similarity is obtained based on the first comparison similarity between the first captured gait features of the specified object and the target gait features. When the first comparison similarity is greater than the third preset similarity, the fifth similarity corresponding to the first comparison similarity is 1. When the first comparison similarity is not greater than the third preset similarity, the fifth similarity corresponding to the first comparison similarity is 0. Among them, the first captured gait features are extracted from the first captured image obtained by capturing the specified object while walking; the target gait features are the gait features of the specified object stored in the target gait feature database; the first captured image parameters include at least one of the image clarity, image pixels, and image resolution of the first captured image.

[0104] Input the first captured image parameters into the first decision tree model to be trained, and output the sixth similarity predicted by the first decision tree model; the value of the sixth similarity is 0 or 1.

[0105] Use the sixth similarity and the fifth similarity to calculate the loss function to obtain the first loss value, and use the first loss value to perform backpropagation training on the first decision tree model to adjust the learnable parameters in the first decision tree model until the prediction accuracy of the first decision tree model reaches the preset accuracy, stop training, and obtain the second decision tree model after training is completed.

[0106] During the application process, when the second captured image of the specified object is captured, determine the second captured image parameters of the second captured image according to the second captured image. The second captured image parameters include at least one of the image clarity, image pixels, and image resolution of the second captured image.

[0107] Input the second captured image parameters into the second decision tree model after training is completed, and output the seventh similarity;

[0108] When the seventh similarity is 1, it means that the captured second captured image is qualified (for example, the image clarity is high enough). At this time, input the second captured image into the gait recognition algorithm to calculate the gait similarity. When the seventh similarity is 0, it means that the captured second captured image is unqualified (for example, the image clarity is unclear). At this time, capture again.

[0109] Embodiment 2:

[0110] Based on the same technical concept, the embodiment of the present application also provides a device for generating a gait recognition algorithm deployment strategy. Figure 3 Shows a schematic structural diagram of generating a gait recognition algorithm deployment strategy provided by the embodiment of the present application, as Figure 3 shown, the device includes:

[0111] An output module 301 is configured to output a decision tree image according to a plurality of parameter combinations used during the training of a trained target decision tree model; the parameter combinations include at least one of acquisition device deployment parameters, shooting environment parameters, video image attribute parameters, and shooting system parameters;

[0112] An extraction module 302 is configured to extract the influence degree of each parameter in the parameter combination on the gait recognition similarity from the decision tree image, and select the top target number of parameters as target parameters in the order of the influence degree of each parameter from large to small;

[0113] A generation module 303 is configured to generate a deployment strategy according to the target parameters, and perform shooting according to the deployment strategy when establishing a target gait feature database.

[0114] Optionally, it further includes:

[0115] An acquisition module is configured to acquire a plurality of training sample data; the training sample data includes the parameter combination and a first similarity corresponding to the parameter combination; the first similarity is obtained according to the comparison similarity between the captured gait feature of a target object and the sample gait feature. When the comparison similarity is greater than a first preset similarity, the first similarity corresponding to the comparison similarity is 1. When the comparison similarity is not greater than the first preset similarity, the first similarity corresponding to the comparison similarity is 0; the sample gait feature is extracted from a first sample image obtained when shooting using a first deployment strategy; the first deployment strategy is generated according to the parameter combination; the captured gait feature is extracted from a second sample image obtained when randomly shooting the target object;

[0116] A division module is configured to divide the training sample data into a training set and a validation set;

[0117] A first input module is configured to input the parameter combination in the training set into at least one initial decision tree model to be trained, and each initial decision tree model respectively outputs a second similarity corresponding to the parameter combination; the value of the second similarity is 0 or 1;

[0118] A first calculation module is configured to, for each initial decision tree model, calculate a loss function using the first similarity and the second similarity output by the initial decision tree model to obtain a loss value, and perform backpropagation training on the initial decision tree model using the loss value to adjust the learnable parameters in the initial decision tree model until the training stops when all the training sample data in the training set are used up, and use the current initial decision tree model as an intermediate decision tree model;

[0119] A second input module, configured to input the parameter combinations in the verification set into each of the intermediate decision tree models, and each of the intermediate decision tree models respectively outputs a third similarity corresponding to the parameter combination; the value of the third similarity is 0 or 1.

[0120] A second calculation module, configured to calculate the accuracy of each of the intermediate decision tree models according to the first similarity and the third similarity output by the intermediate decision tree model.

[0121] A selection module, configured to select, according to the accuracy of each of the intermediate decision tree models, the intermediate decision tree model with the highest accuracy from all the intermediate decision tree models as the target decision tree model.

[0122] Optionally, when the generation module 303 is configured to generate a deployment strategy according to the target parameter and perform shooting according to the deployment strategy when establishing a target gait feature database, it is specifically configured to:

[0123] Obtain a first gait feature and a second gait feature of a test object; the first gait feature is extracted from a first test image obtained by shooting the test object using the deployment strategy; the second gait feature is extracted from a second test image obtained by randomly shooting the test object.

[0124] Calculate a fourth similarity between the first gait feature and the second gait feature, and when the fourth similarity is greater than a second preset similarity, perform shooting according to the deployment strategy when establishing a target gait feature database.

[0125] Optionally, the deployment strategy includes a target shooting duration when shooting a specified object; the device further includes:

[0126] A prompt module, configured to, during the process of shooting according to the deployment strategy for the purpose of establishing a target gait feature database, when the shooting duration when shooting the specified object is less than the target shooting duration, send a prompt message to the specified object and reshoot the specified object; the prompt message is used to prompt the specified object to slow down its walking speed.

[0127] Optionally, the collection device deployment parameters include at least one of the number and pitch angle of the collection devices; the shooting environment parameters include at least one of the complexity and light intensity of the shooting environment; the video image attribute parameters include at least one of the collection duration, shooting angle, sequence length of the captured video, and the image clarity, image resolution, and image pixels of each frame image in the captured video; the shooting system parameters include a frame skipping strategy for selecting images from the video.

[0128] For the specific implementation method steps and principles, refer to the description of Embodiment 1, and details are not elaborated here.

[0129] Embodiment 3:

[0130] Based on the same technical concept, the embodiment of the present application further provides an electronic device. Figure 4 The structural schematic diagram of an electronic device provided by the embodiment of the present application is shown. As Figure 4 shown, the electronic device 400 includes: a processor 401, a memory 402, and a bus 403. The memory stores machine-readable instructions executable by the processor. When the electronic device runs, the processor 401 communicates with the memory 402 through the bus 403, and the processor 401 executes the machine-readable instructions to execute the method steps described in Embodiment 1.

[0131] For the specific implementation method steps and principles, refer to the description of Embodiment 1, and details are not elaborated here.

[0132] Embodiment 4:

[0133] Based on the same technical concept, Embodiment 4 of the present application further provides a computer-readable storage medium. A computer program is stored on the computer-readable storage medium, and when the computer program is run by a processor, it executes the method steps described in Embodiment 1.

[0134] For the specific implementation method steps and principles, refer to the description of Embodiment 1, and details are not elaborated here.

[0135] Those skilled in the art can clearly understand that for the convenience and conciseness of description, the specific working processes of the systems, devices, and units described above can refer to the corresponding processes in the foregoing method embodiments, and details are not elaborated here.

[0136] In the several embodiments provided by the present application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of the units is only a logical function division, and there may be other division methods in actual implementation. For another example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point, the displayed or discussed mutual coupling or direct coupling or communication connection may be through some communication interfaces, and the indirect coupling or communication connection of the devices or units may be in an electrical, mechanical, or other form.

[0137] The unit described as a separation component may or may not be physically separated. The component shown as a unit may or may not be a physical unit, that is, it may be located in one place or may be distributed across multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0138] In addition, in each embodiment of the present application, each functional unit may be integrated in a processing unit, may exist physically alone for each unit, or two or more units may be integrated in one unit.

[0139] If the described function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a non-volatile computer-readable storage medium executable by a processor. Based on such an understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in each embodiment of the present application. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs that can store program codes.

[0140] Finally, it should be noted that: the above-described embodiments are only specific implementation manners of the present application, used to illustrate the technical solutions of the present application, rather than limiting it. The protection scope of the present application is not limited thereto. Although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: any person skilled in the art within the technical scope disclosed by the present application can still modify the technical solutions recorded in the foregoing embodiments, or can easily think of changes, or perform equivalent replacements for some of the technical features; and these modifications, changes, or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for generating a gait recognition algorithm deployment strategy, characterized in that Including: The trained target decision tree model outputs a decision tree image according to multiple parameter combinations used during its training process; The parameter combinations include at least one of acquisition device deployment parameters, shooting environment parameters, video image attribute parameters, and shooting system parameters; Extract the influence degree of each parameter in the parameter combination on the gait recognition similarity from the decision tree image, and select the top target number of parameters in descending order of the influence degree of each parameter as target parameters; Generate a deployment strategy according to the target parameters for shooting according to the deployment strategy when establishing a target gait feature database; The target decision tree model is trained in the following manner: Obtain multiple training sample data; the training sample data includes the parameter combination and the first similarity corresponding to the parameter combination; the first similarity is obtained based on the comparison similarity between the captured gait feature of the target object and the sample gait feature. When the comparison similarity is greater than the first preset similarity, the first similarity corresponding to the comparison similarity is 1. When the comparison similarity is not greater than the first preset similarity, the first similarity corresponding to the comparison similarity is 0; the sample gait feature is extracted from the first sample image obtained when shooting using the first deployment strategy; the first deployment strategy is generated according to the parameter combination; the captured gait feature is extracted from the second sample image obtained when randomly shooting the target object; Divide the training sample data into a training set and a validation set; Input the parameter combinations in the training set into at least one initial decision tree model to be trained, and each initial decision tree model respectively outputs the second similarity corresponding to the parameter combination; the value of the second similarity is 0 or 1; For each initial decision tree model, calculate the loss value using the first similarity and the second similarity output by the initial decision tree model, and use the loss value to perform backpropagation training on the initial decision tree model to adjust the learnable parameters in the initial decision tree model until the training stops when all the training sample data in the training set are used up, and use the current initial decision tree model as an intermediate decision tree model; Input the parameter combinations in the validation set into each intermediate decision tree model, and each intermediate decision tree model respectively outputs the third similarity corresponding to the parameter combination; the value of the third similarity is 0 or 1; For each intermediate decision tree model, calculate the accuracy of the intermediate decision tree model according to the first similarity and the third similarity output by the intermediate decision tree model; Select the one with the highest accuracy from all the intermediate decision tree models as the target decision tree model according to the accuracy of each intermediate decision tree model.

2. The method according to claim 1, wherein The generating a deployment strategy according to the target parameters for shooting according to the deployment strategy when establishing a target gait feature database includes: Obtain the first gait feature and the second gait feature of the test object; the first gait feature is extracted from the first test image obtained by photographing the test object using the deployment strategy; the second gait feature is extracted from the second test image obtained when randomly photographing the test object; Calculate the fourth similarity between the first gait feature and the second gait feature. When the fourth similarity is greater than the second preset similarity, photograph according to the deployment strategy when establishing the target gait feature database.

3. The method according to claim 1, characterized in that, The deployment strategy includes the target shooting duration when photographing a specified object; the method further includes: During the process of photographing according to the deployment strategy for the purpose of establishing the target gait feature database, when the shooting duration when photographing the specified object is less than the target shooting duration, send a prompt message to the specified object and re-photograph the specified object; the prompt message is used to prompt the specified object to slow down its walking speed.

4. The method according to claim 1, wherein The acquisition device deployment parameters include at least one of the number and pitch angle of the acquisition devices; the shooting environment parameters include at least one of the complexity and light intensity of the shooting environment; the video image attribute parameters include at least one of the acquisition duration, shooting angle of view, sequence length of the captured video, and the image clarity, image resolution, and image pixels of each frame image in the captured video; the shooting system parameters include the frame skipping strategy for selecting images from the video.

5. A gait recognition algorithm deployment strategy generation device, characterized in that It includes: An output module, configured to output a decision tree image according to a plurality of parameter combinations used during the training process of the trained target decision tree model; The parameter combinations include at least one of acquisition device deployment parameters, shooting environment parameters, video image attribute parameters, and shooting system parameters; An extraction module, configured to extract the influence degree of each parameter in the parameter combination on the gait recognition similarity from the decision tree image, and select the top target number of parameters in descending order of the influence degree of each parameter as the target parameters; A generation module, configured to generate a deployment strategy according to the target parameters, so as to photograph according to the deployment strategy when establishing the target gait feature database; It further includes: An acquisition module, configured to acquire a plurality of training sample data; the training sample data includes the parameter combination and the first similarity corresponding to the parameter combination; the first similarity is obtained according to the comparison similarity between the captured gait feature of the target object and the sample gait feature. When the comparison similarity is greater than the first preset similarity, the first similarity corresponding to the comparison similarity is 1. When the comparison similarity is not greater than the first preset similarity, the first similarity corresponding to the comparison similarity is 0; the sample gait feature is extracted from the first sample image obtained by photographing using the first deployment strategy; the first deployment strategy is generated according to the parameter combination; the captured gait feature is extracted from the second sample image obtained when randomly photographing the target object; A partitioning module for partitioning the training sample data into a training set and a validation set; A first input module for inputting the parameter combinations in the training set into at least one initial decision tree model to be trained, and each of the initial decision tree models respectively outputs a second similarity corresponding to the parameter combination; the value of the second similarity is 0 or 1; A first calculation module for, for each of the initial decision tree models, calculating a loss function using the first similarity and the second similarity output by the initial decision tree model to obtain a loss value, and performing backpropagation training on the initial decision tree model using the loss value to adjust the learnable parameters in the initial decision tree model until the training stops when all the training sample data in the training set are used up, and taking the current initial decision tree model as an intermediate decision tree model; A second input module for inputting the parameter combinations in the validation set into each of the intermediate decision tree models, and each of the intermediate decision tree models respectively outputs a third similarity corresponding to the parameter combination; the value of the third similarity is 0 or 1; A second calculation module for, for each of the intermediate decision tree models, calculating the accuracy of the intermediate decision tree model according to the first similarity and the third similarity output by the intermediate decision tree model; A selection module for selecting, according to the accuracy of each of the intermediate decision tree models, the intermediate decision tree model with the highest accuracy as the target decision tree model.

6. The device according to claim 5, characterized in that, When the generation module is used to generate a deployment strategy according to the target parameters for shooting according to the deployment strategy when establishing a target gait feature database, it is specifically used for: Obtaining a first gait feature and a second gait feature of a test object; the first gait feature is extracted from a first test image obtained by shooting the test object using the deployment strategy; the second gait feature is extracted from a second test image obtained by randomly shooting the test object; Calculating a fourth similarity between the first gait feature and the second gait feature, and when the fourth similarity is greater than a second preset similarity, shooting according to the deployment strategy when establishing a target gait feature database.

7. An electronic device, characterized in that, Including: A processor, a memory and a bus, the memory stores machine-readable instructions executable by the processor, when the electronic device runs, the processor communicates with the memory through the bus, and when the machine-readable instructions are executed by the processor, the steps of the method according to any one of claims 1 to 4 are executed.

8. A computer-readable storage medium, characterized in that, A computer program is stored on the computer-readable storage medium, and when the computer program is run by the processor, the steps of the method according to any one of claims 1 to 4 are executed.

Images