CN114627310A

CN114627310A - Clothing changing pedestrian weight recognition method, device, equipment and storage medium

Info

Publication number: CN114627310A
Application number: CN202210257785.5A
Authority: CN
Inventors: 陈浩彬; 乔宇; 焦国华
Original assignee: Shenzhen Institute of Advanced Technology of CAS
Current assignee: Shenzhen Institute of Advanced Technology of CAS
Priority date: 2022-03-16
Filing date: 2022-03-16
Publication date: 2022-06-14

Abstract

The invention discloses a method, a device, equipment and a storage medium for identifying the weight of a clothes-changing pedestrian. The method comprises the following steps: acquiring a human body picture to be inquired, a target clothes template picture and a bottom bank pedestrian picture set; acquiring target pedestrian fusion characteristics corresponding to the human body picture to be inquired and the target clothes template picture and basement pedestrian fusion characteristics corresponding to each basement pedestrian picture in the basement pedestrian picture set by adopting a pre-trained clothes-changing pedestrian characteristic identification model, wherein the clothes-changing pedestrian characteristic identification model comprises a biological characteristic identification sub-model, a clothes characteristic identification sub-model and a characteristic fusion sub-model; determining the similarity of the target pedestrian corresponding to the pedestrian pictures of each basement according to the pedestrian fusion characteristics of each basement and the target pedestrian fusion characteristics; and determining the pedestrian picture of the basement with the similarity of the target pedestrian being more than or equal to a preset similarity threshold as the pedestrian picture of the target pedestrian. The invention realizes accurate identification of clothes-changing pedestrians and reduces the manual searching cost.

Description

Clothing changing pedestrian re-identification method, device, equipment and storage medium

Technical Field

The invention relates to the technical field of image recognition, in particular to a clothes-changing pedestrian re-recognition method, device, equipment and storage medium.

Background

The clothing changing pedestrian re-identification technology is that any one picture of a certain target person is given as a query condition, and a picture of the target person changing clothing is found from tens of millions of bottom library pictures. The pedestrian re-identification technology plays a great role in an intelligent security monitoring scene, and can assist tasks such as finding back lost children and tracking personnel.

The common pedestrian re-identification technology mainly focuses on the characteristics of clothes of a target person and utilizes the clothes to 'find a person'. The difficulty of re-identifying the clothes-changed pedestrians is just reflected in that the clothes of the target person are replaced, and the extraction of the characteristics irrelevant to the clothes from the human body picture is difficult because the proportion of the biological characteristics such as the human face and the like in the human body picture is small and the human face is not necessarily visible. Most of the existing clothes-changing pedestrian re-identification methods pay attention to how to extract the edge and body type information of a person, people are found through the body type of the person, but the body contour information is abstract and difficult to extract, and the recognition effect is poor due to the fact that the body contour information is easily interfered by a shooting angle and a shelter and the influence of the change of the posture of a human body.

Disclosure of Invention

The invention provides a clothes-changing pedestrian re-identification method, device, equipment and storage medium, so as to realize accurate identification of clothes-changing pedestrians.

According to an aspect of the present invention, there is provided a method for re-identifying a clothed pedestrian, the method including:

acquiring a human body picture to be inquired, a target clothes template picture and a bottom bank pedestrian picture set;

acquiring target pedestrian fusion characteristics corresponding to the human body picture to be inquired and the target clothes template picture and basement pedestrian fusion characteristics corresponding to each basement pedestrian picture in the basement pedestrian picture set by adopting a pre-trained clothes-changing pedestrian characteristic identification model, wherein the clothes-changing pedestrian characteristic identification model comprises a biological characteristic identification sub-model, a clothes characteristic identification sub-model and a characteristic fusion sub-model;

determining the similarity of the target pedestrian corresponding to the pedestrian pictures of each basement according to the pedestrian fusion characteristics of each basement and the target pedestrian fusion characteristics;

and determining the pedestrian picture of the basement with the similarity of the target pedestrian being more than or equal to a preset similarity threshold as the pedestrian picture of the target pedestrian.

According to another aspect of the present invention, there is provided a clothes-change pedestrian re-recognition apparatus, including:

the data acquisition module is used for acquiring a human body picture to be inquired, a target clothes template picture and a basement pedestrian picture set;

the characteristic extraction module is used for acquiring target pedestrian fusion characteristics corresponding to the human body picture to be inquired and the target clothes template picture and basement pedestrian fusion characteristics corresponding to each basement pedestrian picture in the basement pedestrian picture set by adopting a pre-trained clothes-changing pedestrian characteristic identification model, wherein the clothes-changing pedestrian characteristic identification model comprises a biological characteristic identification sub-model, a clothes characteristic identification sub-model and a characteristic fusion sub-model;

the similarity determining module is used for determining the similarity of the target pedestrian corresponding to the pedestrian pictures of each basement according to the pedestrian fusion characteristics of each basement and the target pedestrian fusion characteristics;

and the target determining module is used for determining the pedestrian pictures in the basement with the target pedestrian similarity greater than or equal to a preset similarity threshold as the target pedestrian pictures.

According to another aspect of the present invention, there is provided an electronic apparatus including:

at least one processor; and

a memory communicatively coupled to the at least one processor; wherein the content of the first and second substances,

the memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor to enable the at least one processor to execute the clothes-changing pedestrian re-identification method according to any embodiment of the invention.

According to another aspect of the present invention, there is provided a computer-readable storage medium storing computer instructions for causing a processor to implement the method for re-identifying a clothed pedestrian according to any one of the embodiments of the present invention when executed.

According to the technical scheme of the embodiment of the invention, a human body picture to be inquired, a target clothes template picture and a bottom warehouse pedestrian picture set are obtained; taking a human body picture to be inquired and a target clothes template picture as input data, and carrying out pedestrian similarity recognition on bottom library pedestrian pictures in a bottom library pedestrian picture set through a pre-trained coat-changing pedestrian re-recognition model to obtain the target pedestrian similarity of each bottom library pedestrian picture, wherein the coat-changing pedestrian re-recognition model comprises a feature recognition model and a similarity recognition model; and determining the pedestrian picture of the basement with the similarity of the target pedestrian being more than or equal to the preset similarity threshold as the target pedestrian picture. The invention solves the problems that in the prior art, the clothes-changing pedestrians are easily influenced by the change of the shooting angle and the human body posture by extracting the figure body shape features, so that the features are difficult to extract and the identification effect is poor.

It should be understood that the statements in this section do not necessarily identify key or critical features of the embodiments of the present invention, nor do they necessarily limit the scope of the invention. Other features of the present invention will become apparent from the following description.

Drawings

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

Fig. 1a is a flowchart of a method for re-identifying a clothes-changing pedestrian according to an embodiment of the present invention;

FIG. 1b is a schematic diagram of model training of a clothes-changing pedestrian re-identification method according to an embodiment of the present invention;

fig. 1c is a diagram illustrating an application effect of a method for re-identifying a clothes-changing pedestrian according to an embodiment of the present invention;

fig. 2 is a flowchart of a clothes-changing pedestrian re-identification method according to a second embodiment of the invention;

fig. 3 is a schematic structural diagram of a clothes-changing pedestrian re-identification device according to a third embodiment of the invention;

fig. 4 is a schematic structural diagram of an electronic device for implementing the clothes-changing pedestrian re-identification method according to the embodiment of the invention.

Detailed Description

In order to make the technical solutions of the present invention better understood, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

It should be noted that the terms "first," "second," "object," and the like in the description and claims of the present invention and in the drawings described above are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used is interchangeable under appropriate circumstances such that the embodiments of the invention described herein are capable of operation in sequences other than those illustrated or described herein. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed, but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.

Example one

Fig. 1a is a flowchart of a clothes-changing pedestrian re-identification method according to an embodiment of the present invention, which is applicable to the case of identifying and positioning clothes-changing pedestrians, and the method can be implemented by a clothes-changing pedestrian re-identification device, which can be implemented in a form of hardware and/or software, and the clothes-changing pedestrian re-identification device can be configured in a computer device.

The application scene of the embodiment can be that lost children and old people are found through monitoring videos, people are located, and the like. For example, in a scene of losing the finding of the child, a family member of the child can provide any picture of the child and a template picture of the same type of clothes worn by the child when the child walks away, and by using the feature recognition model of the pedestrian who changes clothes provided by the embodiment, the picture of the child wearing the specific clothes can be found in the city monitoring picture base; in a scene of tracking a person, the police can provide any one picture of the person and a newly obtained clothes template picture of a witness descriptor, and quickly position the image of the person in a monitoring video.

As shown in fig. 1a, the method comprises:

s110, obtaining a human body picture to be inquired, a target clothes template picture and a bottom bank pedestrian picture set.

The human body picture to be inquired can be understood as any human body picture of a target person needing to be inquired. The target clothes template picture can be understood as a photo or an image of the target clothes worn by the target person needing to be inquired. The pedestrian picture set in the basement can be understood as a set containing all the inquired material pictures. That is to say, the clothes-changing pedestrian re-identification method of the embodiment is to search a picture taken when a target person in a human body picture to be inquired wears a target clothes in a target clothes template picture in the basement pedestrian picture set.

S120, adopting a pre-trained clothes-changing pedestrian feature recognition model to obtain target pedestrian fusion features corresponding to the human body picture to be inquired and the target clothes template picture and bottom bank pedestrian fusion features corresponding to all bottom bank pedestrian pictures in the bottom bank pedestrian picture set. The clothes changing pedestrian feature recognition model comprises a biological feature recognition sub-model, a clothes feature recognition sub-model and a feature fusion sub-model.

The coat-changing pedestrian feature recognition model in the embodiment can be obtained by training a large amount of training data in advance. The coat changing pedestrian feature recognition model can comprise a biological feature recognition sub-model, a coat feature recognition sub-model and a feature fusion sub-model, the biological feature recognition sub-model can be used for obtaining biological features such as appearance outlines of human bodies in pictures, the coat feature recognition sub-model can be used for obtaining coat features of clothes worn by pedestrians in the pictures, and the feature fusion sub-model can be used for fusing the biological features and the coat features to obtain fusion features containing more information of the pedestrians in the pictures.

The target pedestrian fusion feature may be understood as a feature value including a human biometric feature and a target clothing feature worn by the target person. The pedestrian fusion feature in the bottom bank can be understood as a feature value containing the human body biological feature of the pedestrian and the feature of the clothes worn by the pedestrian in the pedestrian picture in the bottom bank.

Specifically, all the basement pedestrian pictures in the set of the human body picture to be inquired, the target clothes template picture and the basement pedestrian picture can be used as input data, a clothes-changing pedestrian feature recognition model is input, and the clothes-changing pedestrian feature recognition model outputs the target pedestrian fusion features corresponding to the human body picture to be inquired and the target clothes template picture and the basement pedestrian fusion features corresponding to each basement pedestrian picture.

Optionally, the training process of the clothing-changing pedestrian feature recognition model in this embodiment may include:

a1, acquiring training data including a human body training picture, a clothes template training picture and a basement pedestrian training picture set, and carrying out pedestrian identification and marking on the basement pedestrian training pictures in the basement pedestrian training picture set in the same training data set according to the human body training picture and the clothes template training picture to obtain standard identification results corresponding to the pedestrian training pictures in the basements.

The training data for training the model in this embodiment includes a query human body training picture, a clothes template training picture, and a bottom bank pedestrian training picture set. Aiming at a model training process, a base-reservoir pedestrian training picture set in a group of training data needs to identify clothes in a clothes template training picture worn by pedestrians in a human body training picture in the group of training data, namely, for a group of training data, a query is made on the human body training picture to include the pedestrian to be searched, the clothes template training picture includes a clothes image worn by the pedestrian to be searched, and the base-reservoir pedestrian training picture set includes a set of material pictures which can be queried. In order to measure the effect of model training, pedestrian identification and marking can be carried out on pedestrian training pictures in the pedestrian training picture set in the bottom bank in advance, and if a pedestrian in a certain pedestrian training picture in the bottom bank is a picture shot when the pedestrian in the human training picture is inquired to wear clothes in the clothes template training picture, the pedestrian training picture in the bottom bank and the inquired human training picture are marked to be included as the same pedestrian.

In this embodiment, a data set including a large amount of real scene clothes-changing pedestrian data may be selected, and training data including a query human body training picture, a clothes template training picture, and a bottom bank pedestrian training picture set may be obtained through operations such as data preprocessing, and used to train a to-be-trained clothes-changing pedestrian feature recognition model. Under the condition that the quantity of the real data is insufficient, large-scale simulation dressing change data can be made as training data through a game simulation means, and the simulation data can comprise different background conditions, shooting angle conditions, weather conditions, shooting distance conditions and the like. The simulation data is low in manufacturing cost and strong in controllability, and can generate large-scale pedestrian data with rich conditions for changing clothes in a short time, so that more and richer learning samples participate in the training process of the model. The comprehensive use of real data and simulation data can enable the model to show higher generalization in real application scenes.

A2, inputting training data into a to-be-trained clothes-changing pedestrian feature recognition model, and obtaining output inquiry biological features, inquiry fusion features, and bottom library biological features and bottom library fusion features corresponding to all bottom library pedestrian training pictures, wherein the to-be-trained clothes-changing pedestrian feature recognition model comprises a to-be-trained biological feature recognition sub-model, a to-be-trained clothes feature recognition sub-model and a to-be-trained feature fusion sub-model.

The to-be-trained clothes-changing pedestrian feature recognition model in the embodiment can comprise a to-be-trained biological feature recognition submodel, a to-be-trained clothes feature recognition submodel and a to-be-trained feature fusion submodel. The biological feature recognition submodel to be trained, the clothes feature recognition submodel to be trained and the feature fusion submodel to be trained can be built by using a network structure such as a transformer or a convolutional neural network, the number of layers of the network can be set according to the training data volume, and the larger the data volume is, the deeper the number of layers can be.

Specifically, the training data can be used as input data, the pedestrian characteristic identification model to be trained for changing clothes is used for identifying, and the inquiry biological characteristics, the inquiry fusion characteristics, the base database biological characteristics and the base database fusion characteristics corresponding to the pedestrian training pictures of each base database corresponding to each group of training data are output.

Further, step a2 can be implemented by the following specific steps:

a21, inputting the human body training picture to be queried into the biological feature recognition submodel to be trained, outputting the intermediate value of the biological feature to be queried and the biological feature to be queried, inputting the pedestrian training picture of each base bank into the biological feature recognition submodel to be trained, and outputting the corresponding intermediate value of the biological feature training of the base bank and the biological feature of the base bank.

The inquiry biological characteristics and the bottom library biological characteristics can be understood as a model output result of the biological characteristic recognition sub-model to be trained; the inquiry of the biological feature intermediate value and the bottom library biological feature training intermediate value can be understood as the output result of the model intermediate layer of the biological feature recognition submodel to be trained.

A22, inputting the clothes template training pictures into the to-be-trained clothes feature recognition submodel, outputting and inquiring clothes feature intermediate values, carrying out clothes detection on each bottom library pedestrian training picture to obtain corresponding bottom library clothes training pictures, inputting each bottom library clothes training picture into the to-be-trained clothes feature recognition submodel, and outputting the corresponding bottom library clothes feature training intermediate values.

The query of the clothes feature intermediate value and the bottom library clothes feature training intermediate value can be understood as an output result of a model intermediate layer of the clothes feature recognition submodel to be trained.

A23, inputting the intermediate values of the biological features and the clothes features into the feature fusion submodel to be trained, outputting the inquiry fusion features, inputting the intermediate values of the biological features of each bottom library and the corresponding intermediate values of the clothes features of the bottom library into the feature fusion submodel to be trained, and outputting the corresponding bottom library fusion features.

In this embodiment, the output result of the model intermediate layer of the biological feature recognition submodel to be trained and the output result of the model intermediate layer of the garment feature recognition submodel to be trained may be fused to obtain the fusion feature. Therefore, the inquiry human body training picture can be input into the biological feature recognition submodel to be trained, the clothes template training picture is input into the clothes feature recognition submodel to be trained, and then the inquiry biological feature intermediate value output by the biological feature recognition submodel to be trained and the inquiry clothes feature intermediate value output by the clothes feature recognition submodel to be trained are input into the feature fusion submodel to be trained, so as to obtain inquiry fusion features; for any bottom bank pedestrian training picture, inputting the bottom bank pedestrian training picture into a biological feature recognition sub-model to be trained, performing clothes detection on the bottom bank pedestrian training picture by using a clothes detection technology, inputting the obtained bottom bank clothes training picture into a clothes feature recognition sub-model to be trained, and further inputting a bottom bank biological feature training intermediate value output by the biological feature recognition sub-model to be trained and a bottom bank clothes feature training intermediate value output by the clothes feature recognition sub-model to be trained into a feature fusion sub-model to be trained to obtain bottom bank fusion features.

In addition, the inquiry biological characteristics and the bottom library biological characteristics are output by using the to-be-trained biological characteristic recognition submodel, and data support is provided for detecting the recognition effect of the to-be-trained biological characteristic recognition submodel.

And A3, performing feature classification on the inquiry biological features, the inquiry fusion features, the biological features of each base library and the fusion features of each base library to obtain a training biological feature recognition result and a training fusion feature recognition result corresponding to the pedestrian training pictures of each base library.

Further, step a3 may be implemented by the following specific steps:

a31, performing feature classification on the inquiry biological features and the biological features of each bottom base by using a first classifier to obtain inquiry biological feature classification results and bottom base biological feature classification results corresponding to the biological features of each bottom base, and determining training biological feature recognition results corresponding to the biological features of each bottom base according to the bottom base biological feature classification results and the inquiry biological feature classification results.

And A32, performing feature classification on the query fusion features and the fusion features of each base library by using a second classifier to obtain a query fusion feature classification result and base library fusion feature classification results corresponding to the fusion features of each base library, and determining training fusion feature recognition results corresponding to the fusion features of each base library according to the query fusion feature classification results and the query fusion feature classification results.

The first classifier and the second classifier can be selected according to specific scenes.

Specifically, the feature classifier may be used to perform feature classification on the biological feature of query, the fusion feature of query, the biological features of each base and the fusion features of each base, and if a certain base and the biological feature of query are classified into the same class, the identification result of the biological feature of training corresponding to the pedestrian training picture of the base may be marked as the same class as the biological feature of query, and similarly, if a certain base and the fusion feature of query are classified into the same class, the identification result of the fusion feature of training corresponding to the pedestrian training picture of the base may be marked as the same class as the fusion feature of query.

And A4, substituting the standard recognition result, the training biological feature recognition result and the training fusion feature recognition result into at least two given loss function expressions respectively to obtain corresponding loss functions.

Specifically, the characteristic recognition effect of the sub-model for recognizing the biological characteristics to be trained can be reflected by comparing the standard recognition result with the training biological characteristic recognition result, and the corresponding loss function can be obtained by combining with the pre-selection of a proper classification loss function expression and the like; by comparing the standard recognition result with the training fusion feature recognition result, the feature recognition effect of the to-be-trained biological feature recognition submodel, the feature recognition effect of the to-be-trained clothes feature recognition submodel and the feature fusion effect of the to-be-trained feature fusion submodel can be reflected, and the corresponding loss function can be obtained by combining with the pre-selected proper classification loss function expression or triple loss function expression and the like.

And A5, reversely propagating the to-be-trained clothes-changing pedestrian feature recognition model through each loss function to obtain the clothes-changing pedestrian feature recognition model.

Specifically, after the loss function is obtained, the pedestrian characteristic identification model for changing clothes to be trained can be subjected to back propagation through the loss function, model parameters are continuously adjusted, and finally the pedestrian characteristic identification model for changing clothes is obtained.

Fig. 1b is a schematic diagram of model training of a clothing-changing pedestrian re-identification method according to an embodiment of the present invention. As shown in fig. 1b, a transform network may be used to construct a biometric feature recognition submodel to be trained, a garment feature recognition submodel to be trained, and a feature fusion submodel to be trained. On one hand, inputting the inquiry human body training picture into the biological feature identifier model to be trained to obtain the inquiry biological feature; inputting the pedestrian training picture of the bottom bank into the biological feature recognition sub-model to be trained, so as to obtain the biological features of the bottom bank; and combining the inquiry biological characteristics and the bottom library biological characteristics with a pre-selected classification loss function expression 1 to obtain a classification loss function 1. On the other hand, inputting the inquired human body training picture into the to-be-trained biological characteristic recognition sub-model, inputting the clothes template training picture into the to-be-trained clothes characteristic recognition sub-model, inputting the inquired biological characteristic intermediate value output by the to-be-trained biological characteristic recognition sub-model and the inquired clothes characteristic intermediate value output by the to-be-trained clothes characteristic recognition sub-model into the to-be-trained characteristic fusion sub-model, and obtaining the output inquiry fusion characteristic; inputting a bottom library pedestrian training picture into a biological feature recognition sub-model to be trained, performing clothes detection on each bottom library pedestrian training picture by using a clothes detector to obtain a bottom library clothes training picture, inputting the bottom library clothes training picture into a clothes feature recognition sub-model to be trained, inputting a bottom library biological feature training intermediate value output by the biological feature recognition sub-model to be trained and a bottom library clothes feature training intermediate value output by the clothes feature recognition sub-model to be trained into a feature fusion sub-model to be trained, and obtaining bottom library fusion features; and combining the query fusion features and the base library fusion features with a pre-selected classification loss function expression 2 and a triple loss function expression to obtain a classification loss function 2 and a triple loss function. The biological characteristic recognition submodel to be trained can be propagated reversely through the classification loss function 1, and the biological characteristic recognition submodel to be trained, the clothes characteristic recognition submodel to be trained and the characteristic fusion submodel to be trained can be propagated reversely through the classification loss function 2 and the triple loss function, so that the clothes changing pedestrian characteristic recognition model is obtained.

And S130, determining the target pedestrian similarity corresponding to the pedestrian pictures of each basement according to the pedestrian fusion characteristics and the target pedestrian fusion characteristics of each basement.

In this embodiment, the target pedestrian similarity may be understood as a similarity between a pedestrian in the bottom library pedestrian picture and a pedestrian in the human body picture to be queried.

Specifically, the feature similarity between each base-bank pedestrian fusion feature and the target pedestrian fusion feature may be calculated, and the calculated similarity may be used as the target pedestrian similarity of the base-bank pedestrian fusion feature. Any similarity calculation method can be selected as the calculation method of the feature similarity, and this embodiment is not limited.

And S140, determining the pedestrian image in the basement with the similarity of the target pedestrian being more than or equal to the preset similarity threshold as the pedestrian image.

Specifically, a similarity threshold may be preset, each target pedestrian similarity is compared with a preset similarity threshold, and if the target pedestrian similarity is greater than or equal to the preset similarity threshold, the corresponding basement pedestrian picture is determined as the target pedestrian picture.

The preset similarity threshold can reflect the feature comparison strictness degree of re-identification of clothes-changing pedestrians, and can be set according to specific application scenes. When the set threshold value is high in value, the feature comparison is considered to be strict, and only when the pedestrian fusion feature of the basement and the target pedestrian fusion feature have high similarity, the pedestrian in the pedestrian picture of the basement corresponding to the pedestrian fusion feature of the basement and the pedestrian in the human body picture to be inquired corresponding to the target pedestrian fusion feature are considered to be the same pedestrian; when the set threshold value is low, the feature contrast can be considered to be relatively loose, and as long as the pedestrian fusion feature of the basement and the target pedestrian fusion feature have certain similarity, the pedestrian in the pedestrian picture of the basement corresponding to the pedestrian fusion feature of the basement and the pedestrian in the human picture to be inquired corresponding to the target pedestrian fusion feature are considered to be the same pedestrian.

For example, fig. 1c is an application effect diagram of a method for re-identifying a clothed changing pedestrian according to an embodiment of the present invention. As shown in fig. 1c, the left side provides the to-be-queried human body picture and the target clothes template picture for the user, and through the coat-changing pedestrian feature recognition model provided in this embodiment, the picture shot after the pedestrian in the to-be-queried human body picture is changed with the clothes in the target clothes template picture, that is, the target pedestrian picture shown in the picture, can be found in the right side basement pedestrian picture set.

Generally, after a certain method or model is used for re-identifying clothes-changing pedestrians, the identification effect of the method or model for re-identifying clothes-changing pedestrians can be measured by counting two indexes, namely mAP and Rank1, and table 1 is an identification index statistical table of several methods and models for re-identifying clothes-changing pedestrians.

TABLE 1 identification index statistical table for various dressing pedestrian re-identification methods and models

Method and model	mAP(％)	Rank1(％)
			ResNet50ibn	1.2	1.0
Vit	5.5	17.5
			Pixel sampling	2.1	11.6
The clothes-changing pedestrian re-identification method provided by the embodiment	25	41.3

As shown in the table, the ResNet50ibn and the Vit methods are difficult to learn and capture the characteristics of the human body which are irrelevant to clothes, so that the indexes of mAP and Rank1 after re-identification detection of clothes changing pedestrians are low; the pixel sampling method utilizes human body analysis as an auxiliary means, and guides the network to focus on body contour features irrelevant to clothes by changing the colors of the clothes in a human body picture, but the effect is not good because the body contour features are abstract, the proportion of the body contour features in the picture is small, the features are difficult to extract, and the overall effect is not good. The clothing change pedestrian re-identification method provided by the embodiment can greatly improve the pedestrian re-identification effect under the clothing change condition, and the mAP and rank1 index values respectively reach 25% and 41.3% in a real test scene, because the method provided by the embodiment inputs a human body picture to be inquired and a target clothes template picture simultaneously when extracting the characteristics, the biological characteristics of the human body and the characteristics of target clothes are respectively extracted through two channels, and then characteristic fusion is carried out, so that the fused characteristics not only include the biological characteristics of the human body, but also include the clothes characteristics of target persons, and the target persons wearing the target clothes can be more easily found in the pedestrian picture set in the basement.

According to the embodiment of the invention, a human body picture to be inquired, a target clothes template picture and a pedestrian picture set in a bottom warehouse are obtained; taking a human body picture to be inquired and a target clothes template picture as input data, and carrying out pedestrian similarity recognition on bottom library pedestrian pictures in a bottom library pedestrian picture set through a pre-trained coat-changing pedestrian re-recognition model to obtain the target pedestrian similarity of each bottom library pedestrian picture, wherein the coat-changing pedestrian re-recognition model comprises a feature recognition model and a similarity recognition model; and determining the pedestrian picture of the basement with the similarity of the target pedestrian being more than or equal to the preset similarity threshold as the target pedestrian picture. The method and the device solve the problems that in the prior art, the clothes changing pedestrian is easily influenced by the change of a shooting angle and a human body posture by extracting the figure body shape characteristic, so that the characteristic is difficult to extract and the identification effect is poor.

Example two

Fig. 2 is a flowchart of a clothes-changing pedestrian re-identification method according to a second embodiment of the present invention, and the embodiment further optimizes the clothes-changing pedestrian re-identification method based on the second embodiment. As shown in fig. 2, the method includes:

s210, obtaining a human body picture to be inquired, a target clothes template picture and a bottom library pedestrian picture set.

And S220, taking the human body picture to be inquired and the target clothes template picture as input data, carrying out feature extraction and feature fusion through a pre-trained clothes-changing pedestrian feature recognition model, and outputting target pedestrian fusion features.

In the embodiment, the biological features such as the posture and the body type of the pedestrian in the human body picture to be inquired and the clothes features in the target clothes template picture can be respectively extracted, and the extracted features are fused to obtain the target pedestrian fusion feature, so that the target pedestrian fusion feature not only contains the biological features of the pedestrian, but also contains the clothes features of the pedestrian after changing clothes, and the accuracy of pedestrian re-identification can be improved.

Optionally, S220 may be implemented by the following specific steps:

s2201, extracting biological characteristics of the human body picture to be inquired by adopting a biological characteristic identification sub-model to obtain a target biological characteristic intermediate value.

Specifically, the biological feature recognition sub-model in the coat-changing pedestrian feature recognition model is mainly used for recognizing the biological features of pedestrians, so that the human body picture to be inquired can be used as input data, feature extraction is performed through the biological feature recognition sub-model, and the network intermediate layer of the biological feature recognition sub-model outputs the target biological feature intermediate value.

S2202, adopting a clothes feature identification sub-model to extract clothes features of the target clothes template picture to obtain a target clothes feature intermediate value.

Specifically, the clothes feature identification submodel in the clothes changing pedestrian feature identification model is mainly used for identifying the clothes features of pedestrians after clothes changing, so that a target clothes template picture can be used as input data, feature extraction is carried out through the clothes feature identification submodel, and a target clothes feature intermediate value is output by a network intermediate layer of the clothes feature identification submodel.

S2203, performing feature fusion on the target biological feature intermediate value and the target clothes feature intermediate value by adopting a feature fusion sub-model to obtain target pedestrian fusion features.

Specifically, the feature fusion sub-model in the coat-changing pedestrian feature recognition model is mainly used for feature fusion of the biological features of the pedestrians and the clothes features of the pedestrians after coat changing, so that the target biological feature intermediate value output by the biological feature recognition sub-model and the target clothes feature intermediate value output by the clothes feature recognition sub-model can be used as input data, the feature fusion sub-model is input, and the feature fusion sub-model performs feature fusion on the target biological feature intermediate value and the target clothes feature intermediate value to output the target pedestrian fusion features.

And S230, performing clothes detection on the pedestrian pictures in the bottom storehouses in the pedestrian picture set to obtain corresponding bottom storehouses clothes template pictures, taking the pedestrian pictures in the bottom storehouses and the corresponding bottom storehouses clothes template pictures as input data, performing feature extraction and feature fusion through the clothes-changing pedestrian feature recognition model, and outputting pedestrian fusion features corresponding to the pedestrian pictures in the bottom storehouses.

In this embodiment, for each pedestrian image in the bottom library, since the clothes worn by the pedestrian in the pedestrian image in the bottom library is the clothes to be identified, the clothes detection technology can be used to detect the clothes area and take the clothes image to obtain the template image of the clothes in the bottom library. When the pedestrian image of the bottom library and the corresponding clothes template image of the bottom library are obtained, the biological characteristics such as the posture and the body type of the pedestrian in the pedestrian image of the bottom library and the clothes characteristics in the clothes template image of the bottom library can be respectively extracted, the extracted characteristics are fused, the pedestrian fusion characteristics of the bottom library are obtained, the pedestrian fusion characteristics of the bottom library comprise the biological characteristics of the pedestrian in the pedestrian image of the bottom library and the clothes characteristics of the pedestrian, and the accuracy of pedestrian re-identification can be improved.

Optionally, for each pedestrian image in the basement, the pedestrian fusion feature in the basement can be extracted through the following specific steps:

s2301, extracting biological features of pedestrian pictures in the bottom bank by adopting a biological feature recognition sub-model, and obtaining a biological feature intermediate value of the bottom bank.

Specifically, similar to extracting the biological features in the human body picture to be inquired, the pedestrian picture in the bottom database can be used as input data, feature extraction is carried out through the biological feature recognition sub-model, and the middle value of the biological features in the bottom database is output by the network middle layer of the biological feature recognition sub-model.

And S2302, adopting a clothes feature recognition sub-model to extract clothes features of the bottom library clothes template picture corresponding to the bottom library pedestrian picture to obtain a bottom library clothes feature intermediate value.

Specifically, similar to the extraction of the clothes features in the target clothes template picture, the bottom library clothes template picture can be used as input data, the features are extracted through a clothes feature identification sub-model, and a network middle layer of the clothes feature identification sub-model outputs a bottom library clothes feature middle value.

And S2303, performing feature fusion on the bottom library biological feature intermediate value and the bottom library clothes feature intermediate value by using a feature fusion sub-model to obtain bottom library pedestrian fusion features.

Specifically, similar to the fusion target biological feature intermediate value and the target clothes feature intermediate value, the bottom library biological feature intermediate value output by the biological feature recognition sub-model and the bottom library clothes feature intermediate value output by the clothes feature recognition sub-model may be input as input data to the feature fusion sub-model, and the feature fusion sub-model performs feature fusion on the bottom library biological feature intermediate value and the bottom library clothes feature intermediate value to output bottom library pedestrian fusion features.

S240, determining the similarity of the target pedestrian corresponding to the pedestrian pictures of each basement according to the pedestrian fusion characteristics of each basement and the target pedestrian fusion characteristics.

And S250, determining the pedestrian image in the basement with the similarity of the target pedestrian being more than or equal to the preset similarity threshold as the pedestrian image.

According to the embodiment of the invention, a human body picture to be inquired, a target clothes template picture and a basement pedestrian picture set are obtained, the human body picture to be inquired and the target clothes template picture are used as input data, feature extraction and feature fusion are carried out through a pre-trained coat-changing pedestrian feature recognition model, and target pedestrian fusion features are output; performing clothes detection on all the pedestrian pictures in the bottom storeroom in the pedestrian picture set of the bottom storeroom to obtain corresponding bottom storeroom clothes template pictures, taking all the pedestrian pictures in the bottom storeroom and the corresponding bottom storeroom clothes template pictures as input data, performing feature extraction and feature fusion through a clothes-changing pedestrian feature recognition model, and outputting pedestrian fusion features corresponding to the pedestrian pictures in the bottom storeroom; according to the pedestrian fusion features and the target pedestrian fusion features of the basements, the corresponding target pedestrian similarity of the pedestrian pictures of the basements is determined, the pedestrian pictures of the basements with the target pedestrian similarity larger than or equal to a preset similarity threshold value are determined to be the target pedestrian pictures, the method can respectively obtain the biological features such as the postures and the shapes of pedestrians and the features of clothes worn by pedestrians, the biological features and the features of clothes worn by pedestrians are fused to obtain the fusion features, pedestrian clothes changing recognition is carried out according to the fusion features, and the recognition accuracy rate of clothes changing pedestrians is improved.

EXAMPLE III

Fig. 3 is a schematic structural diagram of a clothes-changing pedestrian re-identification device according to a third embodiment of the present invention. As shown in fig. 3, the apparatus includes:

and the data acquisition module 310 is used for acquiring a human body picture to be inquired, a target clothes template picture and a basement pedestrian picture set.

The feature extraction module 320 is configured to obtain, by using a pre-trained coat-changing pedestrian feature recognition model, a target pedestrian fusion feature corresponding to the human body picture to be queried and the target clothes template picture, and a basement pedestrian fusion feature corresponding to each basement pedestrian picture in the basement pedestrian picture set, where the coat-changing pedestrian feature recognition model includes a biological feature recognition sub-model, a clothes feature recognition sub-model, and a feature fusion sub-model.

And the similarity determining module 330 is configured to determine the similarity of the target pedestrian corresponding to the pedestrian pictures of each bottom bank according to the pedestrian fusion features of each bottom bank and the target pedestrian fusion features.

And the target determining module 340 is configured to determine the pedestrian image in the basement with the target pedestrian similarity greater than or equal to a preset similarity threshold as the target pedestrian image.

Optionally, the feature extraction module 320 includes:

the target pedestrian fusion feature extraction unit is used for taking the human body picture to be inquired and the target clothes template picture as input data, performing feature extraction and feature fusion through a pre-trained clothes-changing pedestrian feature recognition model, and outputting target pedestrian fusion features;

and the bottom library pedestrian fusion feature extraction unit is used for performing clothes detection on the pedestrian pictures in the bottom library in a centralized manner to obtain corresponding bottom library clothes template pictures, taking the pedestrian pictures in the bottom library and the corresponding bottom library clothes template pictures as input data, performing feature extraction and feature fusion through the clothes-changing pedestrian feature identification model, and outputting the pedestrian fusion features corresponding to the pedestrian pictures in the bottom library.

Optionally, the target pedestrian fusion feature extraction unit is specifically configured to:

adopting the biological characteristic identification submodel to extract biological characteristics of the human body picture to be inquired to obtain a target biological characteristic intermediate value;

adopting the clothes feature identification submodel to extract clothes features of the target clothes template picture to obtain a target clothes feature intermediate value;

and performing feature fusion on the target biological feature intermediate value and the target clothes feature intermediate value by adopting the feature fusion sub-model to obtain target pedestrian fusion features.

Optionally, the basement pedestrian fusion feature extraction unit is specifically configured to:

for each pedestrian picture in the basement, adopting the biological feature recognition sub-model to extract biological features of the pedestrian picture in the basement so as to obtain a biological feature intermediate value in the basement;

adopting the clothes feature identification submodel to extract clothes features of the bottom library clothes template picture corresponding to the bottom library pedestrian picture to obtain a bottom library clothes feature intermediate value;

and performing characteristic fusion on the bottom library biological characteristic intermediate value and the bottom library clothes characteristic intermediate value by adopting the characteristic fusion sub-model to obtain bottom library pedestrian fusion characteristics.

Optionally, the training process of the clothing-changing pedestrian feature recognition model includes:

acquiring training data including a query human body training picture, a clothes template training picture and a bottom bank pedestrian training picture set, and performing pedestrian identification and marking on bottom bank pedestrian training pictures in the bottom bank pedestrian training picture set in the same group of training data according to the query human body training picture and the clothes template training picture to obtain a standard identification result corresponding to each bottom bank pedestrian training picture;

inputting the training data into a to-be-trained clothes-changing pedestrian feature recognition model to obtain output inquiry biological features, inquiry fusion features, and bottom library biological features and bottom library fusion features corresponding to each bottom library pedestrian training picture, wherein the to-be-trained clothes-changing pedestrian feature recognition model comprises a to-be-trained biological feature recognition sub-model, a to-be-trained clothes feature recognition sub-model and a to-be-trained feature fusion sub-model;

obtaining a training biological feature recognition result and a training fusion feature recognition result corresponding to each base bank pedestrian training picture by carrying out feature classification on the inquiry biological feature, the inquiry fusion feature, each base bank biological feature and each base bank fusion feature;

respectively substituting the standard recognition result, the training biological feature recognition result and the training fusion feature recognition result into at least two given loss function expressions to obtain corresponding loss functions;

and carrying out back propagation on the to-be-trained clothes-changing pedestrian feature recognition model through each loss function to obtain the clothes-changing pedestrian feature recognition model.

Optionally, the inputting the training data into the pedestrian feature recognition model to be trained to change clothes to obtain the output inquiry biological features, the inquiry fusion features, and the bottom bank biological features and the bottom bank fusion features corresponding to each pedestrian training picture of the bottom bank includes:

inputting the inquiry human body training picture into the biological feature recognition submodel to be trained, outputting an inquiry biological feature intermediate value and an inquiry biological feature, inputting each bottom bank pedestrian training picture into the biological feature recognition submodel to be trained, and outputting a corresponding bottom bank biological feature training intermediate value and a bottom bank biological feature;

inputting the clothes template training pictures into the to-be-trained clothes feature recognition submodel, outputting and inquiring a clothes feature intermediate value, performing clothes detection on each bottom library pedestrian training picture to obtain a corresponding bottom library clothes training picture, inputting each bottom library clothes training picture into the to-be-trained clothes feature recognition submodel, and outputting a corresponding bottom library clothes feature training intermediate value;

inputting the intermediate values of the biological features to be inquired and the intermediate values of the clothes features to be inquired into the feature fusion submodel to be trained, outputting the inquiry fusion features, inputting the intermediate values of the biological features to be inquired and the corresponding intermediate values of the clothes features to be inquired into the feature fusion submodel to be trained, and outputting the corresponding bottom library fusion features.

Optionally, the obtaining of the training biometric feature recognition result and the training fusion feature recognition result corresponding to each pedestrian training picture of the base database by performing feature classification on the query biometric feature, the query fusion feature, each base database biometric feature and each base database fusion feature includes:

performing feature classification on the inquiry biological features and the biological features of each bottom base by using a first classifier to obtain inquiry biological feature classification results and biological feature classification results of the bottom bases corresponding to the biological features of each bottom base, and determining training biological feature identification results corresponding to the biological features of each bottom base according to the biological feature classification results of each bottom base and the biological feature classification results of the inquiry;

and performing feature classification on the query fusion features and the base library fusion features by using a second classifier to obtain query fusion feature classification results and base library fusion feature classification results corresponding to the base library fusion features, and determining training fusion feature identification results corresponding to the base library fusion features according to the base library fusion feature classification results and the query fusion feature classification results.

The clothes changing pedestrian re-identification device provided by the embodiment of the invention can execute the clothes changing pedestrian re-identification method provided by any embodiment of the invention, and has corresponding functional modules and beneficial effects of the execution method.

Example four

FIG. 4 shows a schematic block diagram of an electronic device 10 that may be used to implement an embodiment of the invention. Electronic devices are intended to represent various forms of digital computers, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframes, and other appropriate computers. The electronic device may also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smart phones, wearable devices (e.g., helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions, are meant to be exemplary only, and are not meant to limit implementations of the inventions described and/or claimed herein.

As shown in fig. 4, the electronic device 10 includes at least one processor 11, and a memory communicatively connected to the at least one processor 11, such as a Read Only Memory (ROM)12, a Random Access Memory (RAM)13, and the like, wherein the memory stores a computer program executable by the at least one processor, and the processor 11 can perform various suitable actions and processes according to the computer program stored in the Read Only Memory (ROM)12 or the computer program loaded from a storage unit 18 into the Random Access Memory (RAM) 13. In the RAM 13, various programs and data necessary for the operation of the electronic apparatus 10 can also be stored. The processor 11, the ROM 12, and the RAM 13 are connected to each other via a bus 14. An input/output (I/O) interface 15 is also connected to bus 14.

A number of components in the electronic device 10 are connected to the I/O interface 15, including: an input unit 16 such as a keyboard, a mouse, or the like; an output unit 17 such as various types of displays, speakers, and the like; a storage unit 18 such as a magnetic disk, an optical disk, or the like; and a communication unit 19 such as a network card, modem, wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information/data with other devices via a computer network such as the internet and/or various telecommunication networks.

Processor 11 may be a variety of general and/or special purpose processing components having processing and computing capabilities. Some examples of processor 11 include, but are not limited to, a Central Processing Unit (CPU), a Graphics Processing Unit (GPU), various specialized Artificial Intelligence (AI) computing chips, various processors running machine learning model algorithms, a Digital Signal Processor (DSP), and any suitable processor, controller, microcontroller, or the like. The processor 11 performs the various methods and processes described above, such as a clothes change pedestrian re-identification method.

In some embodiments, the clothes-change pedestrian re-identification method may be implemented as a computer program tangibly embodied in a computer-readable storage medium, such as the storage unit 18. In some embodiments, part or all of the computer program may be loaded and/or installed onto the electronic device 10 via the ROM 12 and/or the communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the clothes-changing pedestrian re-identification method described above may be performed. Alternatively, in other embodiments, the processor 11 may be configured to perform the clothed pedestrian re-identification method by any other suitable means (e.g., by means of firmware).

Various implementations of the systems and techniques described here above may be implemented in digital electronic circuitry, integrated circuitry, Field Programmable Gate Arrays (FPGAs), Application Specific Integrated Circuits (ASICs), Application Specific Standard Products (ASSPs), system on a chip (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and/or combinations thereof. These various embodiments may include: implemented in one or more computer programs that are executable and/or interpretable on a programmable system including at least one programmable processor, which may be special or general purpose, receiving data and instructions from, and transmitting data and instructions to, a storage system, at least one input device, and at least one output device.

A computer program for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general purpose computer, special purpose computer, or other programmable data processing apparatus, such that the computer programs, when executed by the processor, cause the functions/acts specified in the flowchart and/or block diagram block or blocks to be performed. A computer program can execute entirely on a machine, partly on a machine, as a stand-alone software package partly on a machine and partly on a remote machine or entirely on a remote machine or server.

In the context of the present invention, a computer-readable storage medium may be a tangible medium that can contain, or store a computer program for use by or in connection with an instruction execution system, apparatus, or device. A computer readable storage medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. Alternatively, the computer readable storage medium may be a machine readable signal medium. More specific examples of a machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer diskette, 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 disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

To provide for interaction with a user, the systems and techniques described here can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to a user; and a keyboard and a pointing device (e.g., a mouse or a trackball) by which a user can provide input to the electronic device. Other kinds of devices may also be used to provide for interaction with a user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form, including acoustic, speech, or tactile input.

The systems and techniques described here can be implemented in a computing system that includes a back-end component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front-end component (e.g., a user computer having a graphical user interface or a web browser through which a user can interact with an implementation of the systems and techniques described here), or any combination of such back-end, middleware, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: local Area Networks (LANs), Wide Area Networks (WANs), blockchain networks, and the Internet.

The computing system may include clients and servers. A client and server are generally remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other. The server can be a cloud server, also called a cloud computing server or a cloud host, and is a host product in a cloud computing service system, so that the defects of high management difficulty and weak service expansibility in the traditional physical host and VPS service are overcome.

It should be understood that various forms of the flows shown above may be used, with steps reordered, added, or deleted. For example, the steps described in the present invention may be executed in parallel, sequentially, or in different orders, and are not limited herein as long as the desired results of the technical solution of the present invention can be achieved.

The above-described embodiments should not be construed as limiting the scope of the invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions may be made in accordance with design requirements and other factors. Any modification, equivalent replacement, and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A clothes-changing pedestrian re-identification method is characterized by comprising the following steps:

2. The method according to claim 1, wherein the obtaining of the target pedestrian fusion features corresponding to the human body picture to be queried and the target clothes template picture and the basement pedestrian fusion features corresponding to each basement pedestrian picture in the basement pedestrian picture set by using a pre-trained coat-changing pedestrian feature recognition model comprises:

taking the human body picture to be inquired and the target clothes template picture as input data, performing feature extraction and feature fusion through a pre-trained clothes-changing pedestrian feature recognition model, and outputting target pedestrian fusion features;

and performing clothes detection on all the pedestrian pictures in the basement in the centralized pedestrian pictures to obtain corresponding basement clothes template pictures, taking all the pedestrian pictures in the basement and the corresponding basement clothes template pictures as input data, performing feature extraction and feature fusion through the clothes-changing pedestrian feature recognition model, and outputting the pedestrian fusion features of the basement corresponding to the pedestrian pictures.

3. The method according to claim 2, wherein the step of taking the human body picture to be inquired and the target clothes template picture as input data, performing feature extraction and feature fusion through a pre-trained clothes-changing pedestrian feature recognition model, and outputting target pedestrian fusion features comprises the steps of:

4. The method according to claim 2, wherein the step of taking each basement pedestrian picture and the corresponding basement clothes template picture as input data, performing feature extraction and feature fusion through the clothes-changing pedestrian feature recognition model, and outputting the corresponding basement pedestrian fusion feature of each basement pedestrian picture comprises the steps of:

5. The method according to claim 1, wherein the training process of the clothing-changing pedestrian feature recognition model comprises:

6. The method according to claim 5, wherein the inputting of the training data into the pedestrian feature recognition model for clothing changing to be trained to obtain the output inquiry biological features, the inquiry fusion features, the base biological features and the base fusion features corresponding to each of the base pedestrian training pictures comprises:

inputting the inquiry human body training pictures into the biological feature recognition submodel to be trained, outputting an inquiry biological feature intermediate value and an inquiry biological feature, inputting each bottom bank pedestrian training picture into the biological feature recognition submodel to be trained, and outputting a corresponding bottom bank biological feature training intermediate value and a bottom bank biological feature;

7. The method according to claim 5, wherein the obtaining of the training biometric identification result and the training fusion feature identification result corresponding to each of the base-library pedestrian training pictures by performing feature classification on the query biometric feature, the query fusion feature, each of the base-library biometric features, and each of the base-library fusion features comprises:

performing feature classification on the inquiry biological features and the biological features of each bottom library by using a first classifier to obtain inquiry biological feature classification results and bottom library biological feature classification results corresponding to the biological features of each bottom library, and determining training biological feature identification results corresponding to the biological features of each bottom library according to the biological feature classification results of each bottom library and the inquiry biological feature classification results;

8. A clothes-changing pedestrian weight recognition device is characterized by comprising:

the similarity determining module is used for determining the similarity of the target pedestrian corresponding to the pedestrian pictures of the bottom storeys according to the pedestrian fusion characteristics of the bottom storeys and the target pedestrian fusion characteristics;

9. An electronic device, characterized in that the electronic device comprises:

at least one processor; and

the memory stores a computer program executable by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-7.

10. A computer-readable storage medium storing computer instructions for causing a processor to implement the method for pedestrian re-identification of dressing change of any one of claims 1-7 when executed.