A closed-loop end-to-end pedestrian re-identification background suppression method and system

Abstract

The present invention provides a closed-loop end-to-end pedestrian re-identification background suppression method and system, comprising extracting features of the input original image using the residual convolution block in the first network, and constraining it through a first loss function L1; obtaining a human body mask with features through Mask r-CNN using a mask prediction loss function L mask Correct the accuracy of the mask; multiply the human mask and the original image to obtain a foreground image, use the foreground image as the input of the second network, and introduce the features of the residual convolution block of the first network into the second network, and use the second loss function L2 for constraint. The present invention has achieved competitive performance on multiple pedestrian re-identification data sets.

Description

Technical Field

[0001] The present invention relates to the technical field of computer vision, and in particular to a closed-loop end-to-end pedestrian re-identification background suppression method and system. Background Art

[0002] The goal of person re-identification is to match the same pedestrian in non-overlapping camera views. With the rapid development of surveillance cameras, a large number of surveillance systems are deployed in public places. Therefore, person re-identification technology has received increasing attention in recent years. It can capture specific targets under non-overlapping camera views and has become a key technology in video surveillance applications. In the past few years, person re-identification has achieved advanced performance through deep learning technology.

[0003] Person re-identification is a challenging task due to factors such as posture change, perspective change, illumination intensity change, and background interference. Background interference can be regarded as one of the main factors affecting the performance of person re-identification systems. When two people have similar appearance information and similar background information, misidentification is likely to occur.

[0004] Under different viewing angles, illumination, pedestrian posture, and background clutter will affect the performance of pedestrian re-identification, especially background clutter. The existing background suppression algorithms for pedestrian re-identification mainly use segmentation algorithms or pedestrian parsing algorithms (such as Mask R-CNN or LIP_JPPNet) to pre-process the original image and remove the background to obtain the foreground image. However, the segmentation algorithm or pedestrian parsing algorithm is trained on its corresponding dataset. When the trained algorithm is applied to the pedestrian re-identification image, due to the problem of domain shift, it is impossible to obtain a perfect foreground image, and some important clues (such as backpacks or companions) may even be lost, which will affect the performance of pedestrian re-identification. The segmentation algorithm is trained on the COCO dataset and directly tested on the re-ID dataset. In this way, there are style differences between the data sets, and the clutter in the background cannot be perfectly filtered out, and some key context clues will be lost. The current pedestrian re-identification background suppression algorithms are not end-to-end networks. They all need to use the segmentation algorithm to pre-process the original image first, and then use it as the input of the convolutional neural network. However, these segmentation algorithms do not obtain ideal foreground images, which will reduce the performance of pedestrian re-identification. Summary of the invention

[0005] In order to solve the technical problems in the prior art that the background in the original image cannot be perfectly filtered out using a segmentation algorithm or a pedestrian parsing algorithm, and some key context clues are lost, thereby reducing the performance of pedestrian re-identification, the present invention proposes a closed-loop end-to-end pedestrian re-identification background suppression method and system to solve the above technical problems.

[0006] According to a first aspect of the present invention, a closed-loop end-to-end pedestrian re-identification background suppression method is proposed, comprising:

[0007] S1: Use the residual convolution block in the first network to extract the features of the input original image and constrain it through the first loss function L1;

[0008] S2: The human mask with features obtained through Mask r-CNN uses the mask prediction loss function L mask Correct the accuracy of the mask;

[0009] S3: Multiply the human body mask and the original image to obtain a foreground image, use the foreground image as the input of the second network, and introduce the features of the residual convolution block of the first network into the second network, and use the second loss function L2 for constraint.

[0010] In some specific embodiments, the first network and the second network use ResNet50 as the backbone network.

[0011] In some specific embodiments, the loss function L of the first network and the second network is also included. 12 , the overall loss function expression is L = L1 + L 12 +L2+λL mask , where λ represents the hyperparameter of the balanced loss function.

[0012] In some specific embodiments, the first loss function, the second loss function, and the loss function of the first path network and the second path network combined all include a triplet loss function and a label smoothing regularized cross entropy loss function.

[0013] In some specific embodiments, the formula of the triple loss function is: Where P represents the number of different pedestrian identities, K0 represents the selection of K0 different images for each pedestrian, and for each training image x a , in and x a Select an image x with the largest similarity distance among the images of the same identity p As a positive sample, in and x a Select an image x with the smallest similarity distance among images of different identities n As negative samples, is the threshold set, d(·,·) is defined as the Euclidean distance, batch represents the parameter size of batch processing; the formula of the cross entropy loss function of label smoothing regularization is Where ε∈[0,1], K is the number of pedestrian categories, and p(k) represents the probability of predicting belonging to category k; the formula of the mask prediction loss function is Where B represents the batch size, z n Represents the human mask output, m n Label representing a human mask.

[0014] In some specific embodiments, in S3, the features of the residual convolution block of the first network are introduced into the second network, which is specifically expressed as in Represents channel cascade operation, It indicates the characteristics of the second network after obtaining the first network. represents the characteristics of the first network, Represents the characteristics of the second network.

[0015] According to a second aspect of the present invention, a computer-readable storage medium is provided, on which one or more computer programs are stored. When the one or more computer programs are executed by a computer processor, any of the above methods is implemented.

[0016] According to a third aspect of the present invention, a closed-loop end-to-end pedestrian re-identification background suppression system is proposed, the system comprising:

[0017] The first network module is configured to extract features of the input original image using the residual convolution block in the first network and constrained by the first loss function L1;

[0018] Mask prediction module: Configure the human mask whose features are obtained through Mask r-CNN to use the mask prediction loss function L mask Correct the accuracy of the mask;

[0019] The second network module is configured to multiply the human body mask and the original image to obtain a foreground image, use the foreground image as the input of the second network, and introduce the features of the residual convolution block of the first network into the second network, and use the second loss function L2 for constraint.

[0020] In some specific embodiments, the first network and the second network use ResNet50 as the backbone network.

[0021] In some specific embodiments, the loss function L of the first network and the second network is also included. 12, the overall loss function expression is L = L1 + L 12 +L2+λL mask , where λ represents the hyperparameter of the balanced loss function.

[0022] In some specific embodiments, the first loss function, the second loss function, and the loss function of the first network and the second network combined all include a triplet loss function and a label smoothing regularized cross entropy loss function; the formula of the triplet loss function is: Where P represents the number of different pedestrian identities, K0 represents the selection of K0 different images for each pedestrian, and for each training image x a , in and x a Select an image x with the largest similarity distance among the images of the same identity p As a positive sample, in and x a Select an image x with the smallest similarity distance among images of different identities n As negative samples, is the threshold set, d(·,·) is defined as the Euclidean distance, batch represents the parameter size of batch processing; the formula of the cross entropy loss function of label smoothing regularization is Where ε∈[0,1], K is the number of pedestrian categories, and p(k) represents the probability of predicting belonging to category k; the formula of the mask prediction loss function is Where B represents the batch size, z n Represents the human mask output, m n Label representing a human mask.

[0023] In some specific embodiments, the features of the residual convolution block of the first network are introduced into the second network, which is specifically expressed as in Represents channel cascade operation, It indicates the characteristics of the second network after obtaining the first network. represents the characteristics of the first network, Represents the characteristics of the second network.

[0024] The present invention proposes a closed-loop end-to-end pedestrian re-identification background suppression method and system, which embeds a pedestrian segmentation module into the network to generate a body mask, and uses the gradually corrected body mask to obtain a perfect foreground image to remove the interference of the background on the pedestrian re-identification algorithm. The pedestrian re-identification performance of the second network is fed back to the body mask prediction module to promote the correction of the body mask. At the same time, the correction of the body mask promotes the second network to suppress background clutter. This method and system have achieved competitive performance on multiple pedestrian re-identification data sets. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The accompanying drawings are included to provide a further understanding of the embodiments and are incorporated into and constitute a part of this specification. The accompanying drawings illustrate the embodiments and together with the description are used to explain the principles of the present invention. Other embodiments and many expected advantages of the embodiments will be readily appreciated as they become better understood by reference to the following detailed description. Other features, objects and advantages of the present application will become more apparent by reading the detailed description of the non-limiting embodiments made with reference to the following drawings:

[0026] Figure 1 is a flow chart of a closed-loop end-to-end pedestrian re-identification background suppression method according to an embodiment of the present application;

[0027] Figure 2 It is a closed-loop end-to-end pedestrian re-identification background suppression algorithm framework diagram of a specific embodiment of the present application;

[0028] Figure 3 is a framework diagram of a closed-loop end-to-end pedestrian re-identification background suppression system according to an embodiment of the present application;

[0029] Figure 4 It is a structural diagram of a computer system suitable for implementing an electronic device of an embodiment of the present application. DETAILED DESCRIPTION

[0030] The present application will be further described in detail below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the relevant invention, rather than to limit the invention. It should also be noted that, for ease of description, only the parts related to the relevant invention are shown in the accompanying drawings.

[0031] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0032] According to a closed-loop end-to-end pedestrian re-identification background suppression method according to an embodiment of the present application, Figure 1 FIG. 1 is a flow chart showing a closed-loop end-to-end pedestrian re-identification background suppression method according to an embodiment of the present application. Figure 1 As shown, the method includes:

[0033] S101: extract features of the input original image using the residual convolution block in the first network, and constrain it through the first loss function L1. The first loss function L1 includes a triple loss function and a cross entropy loss function of label smoothing regularization.

[0034] In a specific embodiment, the formula of the triple loss function is: Where P represents the number of different pedestrian identities, K0 represents the selection of K0 different images for each pedestrian, and for each training image x a , in and x a Select an image x with the largest similarity distance among the images of the same identity p As a positive sample, in and x a Select an image x with the smallest similarity distance among images of different identities n As negative samples, is the threshold set for consideration, d(·,· is defined as the Euclidean distance, and batch represents the parameter size of batch processing.

[0035] In some specific embodiments, the formula of the cross entropy loss function of label smoothing regularization is: Where ε∈[0,1], K is the number of pedestrian categories, and p(k) represents the probability of predicting belonging to category k.

[0036] S102: The human mask with features obtained through Mask r-CNN uses the mask prediction loss function L mask Correct the accuracy of the mask. The formula of the mask prediction loss function is Where B represents the batch size, z n Represents the human mask output, m n Label representing a human mask.

[0037] S103: Multiply the human body mask and the original image to obtain a foreground image, use the foreground image as the input of the second network, and introduce the features of the residual convolution block of the first network into the second network, and use the second loss function L2 for constraint. Specifically expressed as in Represents channel cascade operation, It indicates the characteristics of the second network after obtaining the first network. represents the characteristics of the first network, Represents the characteristics of the second network.

[0038] In a specific embodiment, the first network and the second network use ResNet50 as the backbone network. It also includes a loss function L for the combination of the first network and the second network. 12 , the overall loss function expression is L = L1 + L 12 + L2+λL mask , where λ represents the hyperparameter of the balance loss function. The first loss function, the second loss function, and the loss function of the first and second networks combined all include the triplet loss function and the cross entropy loss function with label smoothing regularization. All loss functions are added together to constrain the entire method.

[0039] In a specific embodiment, the second network uses the foreground image as input, and its pedestrian re-identification performance is fed back to the human mask prediction module to supervise the correction of the mask, so that the mask is continuously improved, which in turn acts on the second network to improve the performance of pedestrian re-identification; the first network provides feature information to the second network, and can also provide supervision information to the second network to improve its re-identification performance.

[0040] In a specific embodiment, Figure 2 A closed-loop end-to-end pedestrian re-identification background suppression algorithm framework diagram according to a specific embodiment of the present application is shown, such as Figure 2 As shown in the figure, the whole framework includes a two-way network and a human mask prediction module, forming a multi-task framework, including pedestrian re-identification task and human mask prediction task. The two tasks promote each other. The pedestrian re-identification algorithm promotes the gradual correction of human mask, and the human mask promotes pedestrian re-identification to gradually suppress background clutter. The specific steps of the algorithm include:

[0041] Step S1: ResNet50 is used as the backbone network for the two-way person re-identification task. The network structure is shown in the following figure. Figure 1 The first network uses the original image as input, extracts features through the four residual convolution blocks of ResNet50, and then uses the triplet loss function and the cross entropy loss function of label smoothing regularization for constraints. The triplet loss function and the cross entropy loss function of label smoothing regularization are shown in the following formula:

[0042]

[0043] Where P represents the number of different pedestrian identities, and K0 represents the selection of K0 different images for each pedestrian. For each training image x a , in and x a Select an image x with the largest similarity distance among the images of the same identity p As a positive sample, in and x a Select an image x with the smallest similarity distance among images of different identities n As negative samples, is the threshold set for consideration, d(·,·) is defined as the Euclidean distance, and batch represents the parameter size of batch processing.

[0044] The cross entropy loss function of label smoothing regularization is as follows:

[0045]

[0046] Where ε∈[0,1], K is the number of pedestrian categories, and p(k) represents the probability of predicting belonging to category k.

[0047] Step S2: The feature map obtained by passing the original image through the first residual convolution block in the first ResNet50 is sent to the human mask prediction module. The human mask is predicted through the encoding module and the decoding module. The mask obtained by Mask r-CNN is used as the label. The accuracy of the mask is continuously corrected under the constraint of the mask prediction loss. The mask prediction loss is as follows:

[0048]

[0049] Where B represents the batch size, z n Represents the human mask output, m n Label representing a human mask.

[0050] Step S3: Multiply the mask obtained by the human mask prediction module and the original image to obtain the foreground image as the input of the second network, and then introduce the features of each residual convolution block of the first network into the second network, as shown in the following expression:

[0051]

[0052] in Represents channel cascade operation, It indicates the characteristics of the second network after obtaining the first network. represents the characteristics of the first network, Represents the characteristics of the second network. The loss function of the second network also uses the triple loss function and the cross entropy loss function of label smoothing regularization.

[0053] Step S4: The loss function of the entire framework includes the loss function of the first network, the loss function of the second network, the loss function of the combination of the first and second networks, and the loss function of the human mask prediction module. All loss functions are added together to constrain the entire algorithm. The second network uses the foreground image as input, and its pedestrian re-identification performance is fed back to the human mask prediction module to supervise the correction of the mask. The continuous improvement of the mask reacts to the second network to improve the performance of pedestrian re-identification. The first network provides feature information to the second network, and can also provide supervision information to the second network to improve its re-identification performance. The overall loss function expression is as follows: L = L1 + L 12 +L2+λL mask , where L1, L2, L 12 and L mask They represent the loss functions of the first channel, the second channel, the first and second channel combined, and the human prediction module respectively. λ represents the hyperparameter of the balanced loss function.

[0054] The present invention proposes a closed-loop end-to-end pedestrian re-identification background suppression algorithm, embeds a pedestrian segmentation module into the network to generate a body mask, and uses the gradually corrected body mask to obtain a perfect foreground image to remove the interference of the background on the pedestrian re-identification algorithm. The pedestrian re-identification performance of the second network is fed back to the body mask prediction module to promote the correction of the body mask. At the same time, the correction of the body mask promotes the second network to suppress background clutter. The present invention has achieved competitive performance on multiple pedestrian re-identification data sets. According to the experiments of the inventors of this application, the present invention has achieved competitive performance on multiple pedestrian re-identification data, achieving 95.2% Rank-1 on Market-1501 and 88.7% Rank-1 on DukeMTMC-reID.

[0055] Continue to refer Figure 3 , Figure 3 The framework diagram of a closed-loop end-to-end pedestrian re-identification background suppression system according to an embodiment of the present application is shown. The system specifically includes a first network module 301, a mask prediction module 302, and a second network module 303. The first network module 301 is configured to extract the features of the input original image using the residual convolution block in the first network, and constrain it through the first loss function L1; the mask prediction module 302 is configured to obtain the human body mask of the features through Maskr-CNN using the mask prediction loss function L mask Correct the accuracy of the mask; the second network module 303 is configured to multiply the human body mask and the original image to obtain a foreground image, use the foreground image as the input of the second network, and introduce the features of the residual convolution block of the first network into the second network, and use the second loss function L2 for constraint.

[0056] Reference below Figure 4 , which shows a schematic diagram of the structure of a computer system 400 suitable for implementing an electronic device of an embodiment of the present application. Figure 4 The electronic device shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0057] like Figure 4As shown, the computer system 400 includes a central processing unit (CPU) 401, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 402 or a program loaded from a storage part 408 into a random access memory (RAM) 403. In the RAM 403, various programs and data required for the operation of the system 400 are also stored. The CPU 401, the ROM 402, and the RAM 403 are connected to each other via a bus 404. An input / output (I / O) interface 405 is also connected to the bus 404.

[0058] The following components are connected to the I / O interface 405: an input section 406 including a keyboard, a mouse, etc.; an output section 407 including a liquid crystal display (LCD), etc. and a speaker, etc.; a storage section 408 including a hard disk, etc.; and a communication section 409 including a network interface card such as a LAN card, a modem, etc. The communication section 409 performs communication processing via a network such as the Internet. A drive 410 is also connected to the I / O interface 405 as needed. A removable medium 411, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 410 as needed, so that a computer program read therefrom is installed into the storage section 408 as needed.

[0059] In particular, according to an embodiment of the present disclosure, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes a computer program carried on a computer-readable storage medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 409, and / or installed from the removable medium 411. When the computer program is executed by the central processing unit (CPU) 401, the above functions defined in the method of the present application are executed. It should be noted that the computer-readable storage medium of the present application can be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium can be, for example, - but not limited to - a system, device or device of electricity, magnetism, light, electromagnetic, infrared, or semiconductor, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to, an electrical connection with one or more conductors, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present application, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, device, or device. In the present application, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, in which a computer-readable program code is carried. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. A computer-readable signal medium may also be any computer-readable storage medium other than a computer-readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, device, or device. The program code contained on the computer-readable storage medium may be transmitted using any appropriate medium, including but not limited to: wireless, wireline, optical cable, RF, etc., or any suitable combination of the foregoing.

[0060] Computer program code for performing the operations of the present application may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages, such as Java, Smalltalk, C++, and conventional procedural programming languages, such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0061] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present application. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0062] The modules involved in the embodiments of the present application may be implemented by software or by hardware.

[0063] As another aspect, the present application also provides a computer-readable storage medium, which may be included in the electronic device described in the above embodiment; or it may exist independently without being assembled into the electronic device. The above computer-readable storage medium carries one or more programs, and when the above one or more programs are executed by the electronic device, the electronic device: uses the residual convolution block in the first network to extract the features of the input original image, and constrains it through the first loss function L1; obtains the human body mask of the features through Mask r-CNN using the mask prediction loss function L maskCorrect the accuracy of the mask; multiply the human body mask and the original image to obtain the foreground image, use the foreground image as the input of the second network, and introduce the features of the residual convolution block of the first network into the second network, and use the second loss function L2 for constraint.

[0064] The above description is only a preferred embodiment of the present application and an explanation of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the present application is not limited to the technical solution formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above invention concept. For example, the above features are replaced with the technical features with similar functions disclosed in this application (but not limited to) by each other to form a technical solution.

Claims

1. A closed-loop end-to-end person re-identification background suppression method, characterized in that: include: S1: Use the residual convolution block in the first network to extract the features of the input original image and use the first loss function To impose restraints; S2: Obtain the human mask of the above features through Mask r-CNN and use the mask prediction loss function Correct the accuracy of the mask; S3: multiply the human body mask and the original image to obtain a foreground image, use the foreground image as the input of the second network, and introduce the features of the residual convolution block of the first network into the second network, using the second loss function To impose restraints; The first network and the second network use ResNet50 as the backbone network; and also include a loss function for the first network and the second network. , the overall loss function expression is ,in, represents a hyperparameter of a balanced loss function; the first loss function, the second loss function, and the loss function of the first network and the second network combined all include a triplet loss function and a label smoothing regularized cross entropy loss function; the formula of the triplet loss function is: ,in represents the number of different pedestrian identities, Select for each pedestrian different images, for each training image , in and Select the image with the largest similarity distance among the images of the same identity As positive samples, Select an image with the smallest similarity distance among images of different identities As negative samples, The threshold value is considered to be Defined as the Euclidean distance, represents the parameter size of batch processing; the formula of the cross entropy loss function of the label smoothing regularization is ,in , is the number of pedestrian categories, Represents the prediction belongs to the category The probability of; the formula of the mask prediction loss function is , where B represents the batch size, represents the human mask output, The label representing the human body mask; in S3, the features of the residual convolution block of the first network are introduced into the second network, which is specifically expressed as ,in Represents channel cascade operation, represents the characteristics of the second network after obtaining the first network, represents the characteristics of the first network, representing the characteristics of the second network; The pedestrian re-identification performance of the second network is fed back to the human mask prediction module to promote the correction of the human mask. At the same time, the correction of the human mask promotes the second network to suppress background clutter.

2. A computer-readable storage medium having one or more computer programs stored thereon, characterized in that: When the one or more computer programs are executed by a computer processor, the method of claim 1 is implemented.

3. A closed-loop end-to-end person re-identification background suppression system, characterized in that: The system comprises: The first network module is configured to use the residual convolution block in the first network to extract the features of the input original image and pass the first loss function To impose restraints; Mask prediction module: configured to obtain the human mask of the features through Mask r-CNN using the mask prediction loss function Correct the accuracy of the mask; The second network module is configured to multiply the human body mask and the original image to obtain a foreground image, use the foreground image as the input of the second network, and introduce the features of the residual convolution block of the first network into the second network, using the second loss function To impose restraints; The first network and the second network use ResNet50 as the backbone network; and also include a loss function for the first network and the second network. , the overall loss function expression is ,in, represents a hyperparameter of a balanced loss function; the first loss function, the second loss function, and the loss function of the first network and the second network combined all include a triplet loss function and a label smoothing regularized cross entropy loss function; the formula of the triplet loss function is: ,in represents the number of different pedestrian identities, Select for each pedestrian different images, for each training image , in and Select the image with the largest similarity distance among the images of the same identity As positive samples, Select an image with the smallest similarity distance among images of different identities As negative samples, The threshold value is considered to be Defined as the Euclidean distance, represents the parameter size of batch processing; the formula of the cross entropy loss function of the label smoothing regularization is ,in , is the number of pedestrian categories, Represents the prediction belongs to the category The probability of; the formula of the mask prediction loss function is , where B represents the batch size, represents the human mask output, The label representing the human body mask; in the second network module, the features of the residual convolution block of the first network are introduced into the second network, which is specifically expressed as ,in Represents channel cascade operation, represents the characteristics of the second network after obtaining the first network, represents the characteristics of the first network, representing the characteristics of the second network; The pedestrian re-identification performance of the second network is fed back to the human mask prediction module to promote the correction of the human mask. At the same time, the correction of the human mask promotes the second network to suppress background clutter.

Images