Pedestrian re-identification method, device, electronic device and computer-readable medium

Abstract

The embodiments of the present disclosure disclose a pedestrian re-identification method, device, electronic device and computer-readable medium. A specific implementation of the method includes: acquiring pedestrian images taken by at least two cameras to obtain a pedestrian image group; performing image preprocessing on each pedestrian image in the above pedestrian image group to obtain a processed pedestrian image group; performing pedestrian recognition on each processed pedestrian image in the above processed pedestrian image group to generate a pedestrian recognition information group set; performing information association processing on each pedestrian recognition information in each pedestrian recognition information group in the above pedestrian recognition information group set to generate an associated pedestrian recognition information set; based on the associated pedestrian recognition information set, performing pedestrian re-identification on the above processed pedestrian images to obtain a pedestrian re-identification result. This implementation can improve the robustness of recognition of occlusion and posture changes in pedestrian images.

Description

Technical Field

[0001] Embodiments of the present disclosure relate to the field of computer technology, and more particularly to a method, device, electronic device, and computer-readable medium for pedestrian re-identification. Background Art

[0002] Person Re-identification (Re-ID) is a computer vision task that aims to match the same individual across different cameras. It has broad application prospects in the fields of intelligent video surveillance systems and personnel behavior analysis. The main purpose of this task is to extract robust person feature representations from different camera perspectives, different lighting conditions, occlusions, and posture changes to achieve accurate individual matching. At present, when performing person re-identification, the methods usually used are: global feature-based methods, which directly extract global features from the entire person image; or local feature-based methods, which divide the person image into strips or grids, extract local features from each area, and finally splice them into overall features.

[0003] However, in practice, it is found that when the above method is used for person re-identification, the following technical problems often occur:

[0004] First, the method based on global features is not very robust to occlusion and posture changes;

[0005] Second, the method based on local features will lead to too high feature dimension and increase computational complexity;

[0006] Third, if the person in the picture is occluded, the disappeared body parts will affect the corresponding feature channels, making it difficult for the model training to converge, thereby affecting the training of other channels, and thus making it difficult for the trained model to better identify the occluded person.

[0007] The above information disclosed in this Background section is only for enhancement of understanding of the background of the inventive concept and therefore it may contain information that does not form the prior art that is already known in this country to a person of ordinary skill in the art. Summary of the invention

[0008] The content of this disclosure is used to introduce concepts in a brief form, which will be described in detail in the detailed implementation section below. The content of this disclosure is not intended to identify the key features or essential features of the technical solution claimed for protection, nor is it intended to limit the scope of the technical solution claimed for protection.

[0009] Some embodiments of the present disclosure propose a pedestrian re-identification method, an apparatus, an electronic device, and a computer-readable medium to solve one or more of the technical problems mentioned in the above background technology section.

[0010] In a first aspect, some embodiments of the present disclosure provide a method for pedestrian re-identification, which includes: acquiring pedestrian images taken by at least two cameras to obtain a pedestrian image group, wherein each pedestrian image in the above pedestrian image group is taken by different cameras at the same time; performing image preprocessing on each pedestrian image in the above pedestrian image group to obtain a processed pedestrian image group; performing pedestrian recognition on each processed pedestrian image in the above processed pedestrian image group to generate a pedestrian recognition information group set, wherein each pedestrian recognition information group in the above pedestrian recognition information group set corresponds to a pedestrian, and each pedestrian recognition information in each pedestrian recognition information group is obtained by identifying different processed pedestrian images; performing information association processing on each pedestrian recognition information in each pedestrian recognition information group in the above pedestrian recognition information group set to generate an associated pedestrian recognition information set; based on the associated pedestrian recognition information set, performing pedestrian re-identification on the above processed pedestrian images to obtain a pedestrian re-identification result.

[0011] In a second aspect, some embodiments of the present disclosure provide a pedestrian re-identification device, the device comprising: an acquisition unit, configured to acquire pedestrian images taken by at least two cameras to obtain a pedestrian image group, wherein each pedestrian image in the above pedestrian image group is taken by different cameras at the same time; a preprocessing unit, configured to perform image preprocessing on each pedestrian image in the above pedestrian image group to obtain a processed pedestrian image group; a first recognition unit, configured to perform pedestrian recognition on each processed pedestrian image in the above processed pedestrian image group to generate a pedestrian recognition information group A set, wherein each pedestrian identification information group in the above pedestrian identification information group set corresponds to a pedestrian, and each pedestrian identification information in each pedestrian identification information group is identified from different processed pedestrian images; an association processing unit is configured to perform information association processing on each pedestrian identification information in each pedestrian identification information group in the above pedestrian identification information group set to generate an associated pedestrian identification information set; a second identification unit is configured to perform pedestrian re-identification on the above processed pedestrian image based on the above associated pedestrian identification information set to obtain a pedestrian re-identification result.

[0012] In a third aspect, some embodiments of the present disclosure provide an electronic device comprising: one or more processors; a storage device on which one or more programs are stored, and when the one or more programs are executed by the one or more processors, the one or more processors implement the method described in any implementation manner of the above-mentioned first aspect.

[0013] In a fourth aspect, some embodiments of the present disclosure provide a computer-readable medium having a computer program stored thereon, wherein when the program is executed by a processor, the method described in any implementation manner of the above-mentioned first aspect is implemented.

[0014] The above-mentioned embodiments of the present disclosure have the following beneficial effects: through the pedestrian re-identification method of some embodiments of the present disclosure, the robustness of identifying occlusion and posture changes in pedestrian images can be improved. Specifically, the reason for the less robustness to occlusion and posture changes is that the global feature method directly extracts global features from the entire person image. Based on this, the pedestrian re-identification method of some embodiments of the present disclosure first obtains pedestrian images taken by at least two cameras to obtain a pedestrian image group. Among them, each pedestrian image in the above pedestrian image group is taken by different cameras at the same time. Then, each pedestrian image in the above pedestrian image group is image preprocessed to obtain a processed pedestrian image group. After that, pedestrian recognition is performed on each processed pedestrian image in the above processed pedestrian image group to generate a pedestrian recognition information group set. Among them, each pedestrian recognition information group in the above pedestrian recognition information group set corresponds to a pedestrian, and each pedestrian recognition information in each pedestrian recognition information group is obtained by identifying different processed pedestrian images. Here, pedestrian recognition can be used to preliminarily detect pedestrian information in the image. Next, information association processing is performed on each of the pedestrian recognition information in each pedestrian recognition information group in the above-mentioned pedestrian recognition information group set to generate an associated pedestrian recognition information set. Here, through the association processing, the features in multiple images can be better combined to highlight the features of occluded people and people with posture changes in the image. This facilitates feature extraction. Finally, based on the above-mentioned associated pedestrian recognition information set, pedestrian re-identification is performed on the above-mentioned processed pedestrian image to obtain a pedestrian re-identification result. Therefore, through pedestrian re-identification, the pedestrian information in the image can be re-identified on the basis of facilitating the extraction of the features of occluded people and people with posture changes in the image. Furthermore, the robustness of the recognition of occlusion and posture changes in pedestrian images can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] The above and other features, advantages and aspects of the embodiments of the present disclosure will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. Throughout the accompanying drawings, the same or similar reference numerals represent the same or similar elements. It should be understood that the drawings are schematic and that components and elements are not necessarily drawn to scale.

[0016] Figure 1 is a flowchart of some embodiments of the pedestrian re-identification method according to the present disclosure;

[0017] Figure 2 is a schematic structural diagram of some embodiments of a pedestrian re-identification device according to the present disclosure;

[0018] Figure 3 It is a schematic diagram of the structure of an electronic device suitable for implementing some embodiments of the present disclosure. DETAILED DESCRIPTION

[0019] Embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although certain embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as being limited to the embodiments set forth herein. On the contrary, these embodiments are provided to provide a more thorough and complete understanding of the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are only for exemplary purposes and are not intended to limit the scope of protection of the present disclosure.

[0020] It should also be noted that, for ease of description, only the parts related to the invention are shown in the drawings. In the absence of conflict, the embodiments and features in the embodiments of the present disclosure can be combined with each other.

[0021] It should be noted that the concepts such as "first" and "second" mentioned in the present disclosure are only used to distinguish different devices, modules or units, and are not used to limit the order or interdependence of the functions performed by these devices, modules or units.

[0022] It should be noted that the modifications of "one" and "plurality" mentioned in the present disclosure are illustrative rather than restrictive, and those skilled in the art should understand that unless otherwise clearly indicated in the context, it should be understood as "one or more".

[0023] The names of the messages or information exchanged between multiple devices in the embodiments of the present disclosure are only used for illustrative purposes and are not used to limit the scope of these messages or information.

[0024] The present disclosure will be described in detail below with reference to the accompanying drawings and in conjunction with embodiments.

[0025] Figure 1 A process 100 of some embodiments of a pedestrian re-identification method according to the present disclosure is shown. The pedestrian re-identification method comprises the following steps:

[0026] Step 101: Obtain pedestrian images taken by at least two cameras to obtain a pedestrian image group.

[0027] In some embodiments, the execution subject of the pedestrian re-identification method can obtain pedestrian images taken by at least two cameras in a wired or wireless manner to obtain a pedestrian image group. Among them, each pedestrian image in the above pedestrian image group can be taken by different cameras at the same time. Secondly, the positions of different cameras can be different, and there are overlapping areas in the shooting angles. Each pedestrian image can correspond to one camera.

[0028] It should be noted that the above-mentioned wireless connection methods may include but are not limited to 3G / 4G connection, WiFi connection, Bluetooth connection, WiMAX connection, Zigbee connection, UWB (ultra wideband) connection, and other wireless connection methods currently known or to be developed in the future.

[0029] Step 102 : performing image preprocessing on each pedestrian image in the pedestrian image group to obtain a processed pedestrian image group.

[0030] In some embodiments, the execution entity may perform image preprocessing on each pedestrian image in the pedestrian image group to obtain a processed pedestrian image group.

[0031] In some optional implementations of some embodiments, the execution subject performs image preprocessing on each pedestrian image in the pedestrian image group to obtain a processed pedestrian image group, which may include the following steps:

[0032] The first step is to perform rough person detection on each pedestrian image in the above pedestrian image group to obtain a rough person detection information group. Among them, each rough person detection information in the above rough person detection information group can indicate whether there is a pedestrian in the pedestrian image. Here, a preset rough person detection model can be used to perform rough person detection on each pedestrian image in the above pedestrian image group to obtain a rough person detection information group. Secondly, each rough person detection information corresponds to a pedestrian image. If a pedestrian is detected, the rough person detection information can include a detection frame for each pedestrian in the image to indicate the presence of a pedestrian in the image.

[0033] As an example, the rough person detection model may include but is not limited to at least one of the following: YOLO-v3 (You Only Look Once-Version3) algorithm, Faster Rcnn (faster regional proposal networks) fast regional convolutional network, etc.

[0034] In the second step, in response to determining that each piece of rough person detection information in the rough person detection information group is information indicating the existence of pedestrians, each pedestrian image in the pedestrian image group is grayscaled to generate a pedestrian grayscale image group.

[0035] The third step is to adjust the contrast of each pedestrian grayscale image in the pedestrian grayscale image group to obtain an adjusted pedestrian image group. Since the camera positions are different, a contrast adjustment ratio can be pre-set for the camera corresponding to each pedestrian grayscale image. Then, the contrast adjustment ratio corresponding to each pedestrian grayscale image can be used to perform contrast adjustment to obtain an adjusted pedestrian image group.

[0036] The fourth step is to perform grayscale slicing on each adjusted pedestrian image in the adjusted pedestrian image group based on the rough detection information group of the task to obtain a processed pedestrian image group. The grayscale slicing can distinguish the background area and the pedestrian area in the pedestrian image to preliminarily remove the background area.

[0037] In practice, by removing the background area, the features of the pedestrian area in the image can be highlighted, thereby facilitating feature extraction, which can be used to improve the accuracy of pedestrian re-identification.

[0038] Step 103 , performing pedestrian recognition on each processed pedestrian image in the processed pedestrian image group to generate a pedestrian recognition information group set.

[0039] In some embodiments, the execution subject may perform pedestrian recognition on each processed pedestrian image in the processed pedestrian image group to generate a pedestrian recognition information group set. Each pedestrian recognition information group in the pedestrian recognition information group set may correspond to a pedestrian. Each pedestrian recognition information in each pedestrian recognition information group may be obtained by recognition from different processed pedestrian images.

[0040] In some optional implementations of some embodiments, the execution subject performs pedestrian recognition on each processed pedestrian image in the processed pedestrian image group to generate a pedestrian recognition information group set, which may include the following steps:

[0041] For each processed pedestrian image in the above processed pedestrian image group, the following recognition steps are performed to generate a pedestrian recognition information group in the pedestrian recognition information group set:

[0042] In the first step, the processed pedestrian image is input into the backbone network of the preset pedestrian re-identification model to extract features and obtain an initial feature map of the image. The pedestrian re-identification model may also include a branch network module. The branch network module may include: a person identity branch network, a human body key point branch network, and a person attribute branch network.

[0043] Specifically, the backbone network may include: multiple convolutional layers and multiple pooling layers. For example, the input image has 3 channels and a size of 224×224. Then, each convolutional layer and pooling layer input to the backbone network can output a feature map with a size of 7×7 and 1024 channels.

[0044] In the second step, the initial feature map of the image is input into the above-mentioned person identity branch network to output the person identity classification information corresponding to each pedestrian in the above-mentioned processed pedestrian image. Among them, the person identity branch network can be a fully connected layer, which is used to classify the input initial feature map of the image to obtain the person identity classification information corresponding to each person. Here, the person identity category information can be the label of the person in the image. For example, the labels corresponding to three people can be 1, 2, and 3.

[0045] The third step is to input the above-mentioned initial feature map of the image into the above-mentioned human key point branch network to output a set of human key point heat map groups corresponding to each pedestrian in the above-mentioned processed pedestrian image. Among them, first, the channels of the initial feature map of the image can be split into a first preset number of groups. Here, the first preset number can be 17, corresponding to the number of human key points. For example, 1024 channels are divided into 17 groups. Then, the feature channels of each group can be input into the deconvolution layer of the human key point branch network for upsampling and decoding to obtain the decoded feature map. After that, the decoded feature map can be convolved 1×1 to generate the above-mentioned first preset number of predicted key point heat maps to obtain a set of human key point heat map groups. Here, each human key point heat map group can correspond to a heat map of different key point parts of a pedestrian.

[0046] Restore to 224

[0047] The fourth step is to determine the human key point identification information group corresponding to the above human key point heat map group. Among them, each human key point identification information in the above human key point identification information group includes a human key point coordinate group and a corresponding key point label group. Secondly, the thermal center point coordinates of each predicted key point heat map in the predicted key point heat map group can be determined as the human key point coordinates to obtain the human key point coordinate group. Here, the channel of each predicted key point heat map can be determined as the key point label.

[0048] The fifth step is to input the above-mentioned initial feature map of the image into the above-mentioned character attribute branch network to output the character attribute classification information corresponding to each pedestrian in the above-mentioned processed pedestrian image. Among them, first, the feature channels of the above-mentioned initial feature map of the image can be split into a second preset number of groups. Here, the second preset number can be the number of character attributes. Then, the feature channels of each group can be input into the character attribute branch network to output the character attribute classification information corresponding to each pedestrian in the above-mentioned processed pedestrian image. Secondly, the character attribute classification information can include at least one of the following: upper clothing attributes, lower clothing attributes, character gender attributes, character action attributes, etc.

[0049] In the sixth step, the person identity classification information, the human body key point recognition information group and the person attribute classification information corresponding to each pedestrian in the above processed pedestrian image are determined as pedestrian recognition information to obtain a pedestrian recognition information group.

[0050] Optionally, the above person re-identification model is trained and generated by the following steps:

[0051] The first step is to obtain a sample image and a corresponding sample label, wherein the sample image may include at least one pedestrian. The sample label may include a pedestrian label group corresponding to each pedestrian. Secondly, each label may include a person identity sample label, a human key point sample feature map group and a person sample attribute label corresponding to each pedestrian in the sample image.

[0052] In the second step, the sample image is input into the initial backbone network of the initial person re-identification model to obtain the extracted person feature map. The initial person re-identification model may further include: an initial person identity branch network, an initial human body key point branch network, and an initial person attribute branch network. Here, the initial person identity branch network may be a fully connected layer.

[0053] In the third step, the extracted pedestrian feature map is respectively input into the initial person identity branch network, the initial human body key point branch network and the initial person attribute branch network to obtain a person identity classification identification group, a human body key point heat map group set and a person attribute classification identification group set. Among them, the human body key point heat map in the human body key point heat map group set can be a heat map centered on the human body key point. Each person identity classification identification, human body key point heat map group and person attribute classification identification group can correspond to a pedestrian.

[0054] The fourth step is to determine the person identity loss value between each person identity classification identifier in the above person identity classification identifier group and the above pedestrian label group. First, the cross entropy loss value between each person identity classification identifier in the above person identity classification identifier group and the above pedestrian label group can be determined by the cross entropy loss function. Then, the triplet loss function in metric learning is used to impose constraints on the distance between the positive sample pair and the negative sample pair, and the triplet loss value is calculated. Finally, the sum of the cross entropy loss value and the triplet loss value can be determined as the person identity loss value.

[0055] In practice, positive and negative sample pairs (positive sample pairs and negative sample pairs) refer to the image pairs used to train the model, where the positive sample pairs represent the image pairs of the same pedestrian under different cameras or time periods, while the negative sample pairs represent the image pairs of different pedestrians. By constructing positive and negative sample pairs, the model can learn how to distinguish the difference between the same pedestrian and different pedestrians, thereby achieving the task of pedestrian re-identification. In addition, the triplet loss function is a loss function used to measure the similarity between image pairs. During the training process, it imposes a smaller loss on the positive sample pairs and a larger loss on the negative sample pairs, so as to encourage the model to learn an effective similarity measure, thereby improving the accuracy and robustness of pedestrian re-identification.

[0056] The fifth step is to determine the character key point loss value between the above-mentioned human key point heat map group set and the above-mentioned pedestrian label group. Among them, the heat map loss value between each human key point heat map group in the above-mentioned human key point heat map group set and the pedestrian label corresponding to the same pedestrian in the above-mentioned pedestrian label group can be determined by the L1 loss function. Finally, the sum of the heat map loss values ​​corresponding to each human key point heat map group can be determined as the character key point loss value. Here, the pedestrian label can include a sample Gaussian heat map corresponding to each human key point in the sample image, and the key point can be pre-marked.

[0057] The sixth step is to determine the character attribute loss value between the character attribute classification identification group set and the pedestrian label group. First, the attribute loss value between each character attribute classification identification group in the character attribute classification identification group set and the pedestrian label corresponding to the same pedestrian in the pedestrian label group can be determined by a cross entropy loss function. Finally, the sum of the attribute loss values ​​corresponding to each character attribute classification identification group can be determined as the character attribute loss value. Here, the pedestrian label can also include a sample attribute label corresponding to each pedestrian in the sample image.

[0058] In the seventh step, the weighted sum of the above-mentioned character identity loss value, the above-mentioned character key point loss value and the above-mentioned character attribute loss value is determined as the total loss value. Among them, the above-mentioned character identity loss value, the above-mentioned character key point loss value and the above-mentioned character attribute loss value can be weighted and summed using two preset weights to obtain the total loss value. Here, in the weighted summation process, the loss value of the character identity loss can be 1. The above-mentioned two weights are the weights of the above-mentioned character key point loss value and the above-mentioned character attribute loss value, respectively. Specifically, the above-mentioned two weights are hyperparameters for balancing the three branches.

[0059] In step 8, in response to determining that the total loss value is less than or equal to the preset loss threshold, and the training of the sample image satisfies the preset training end condition, the initial pedestrian re-identification model is determined as a trained pedestrian re-identification model. The training of the sample image satisfies the preset training end condition, which may be that the training round in which the sample image is located is greater than the preset training round. It may also be that the sample image is the last sample image in the sample image set used for training. Here, the end condition of the model training is not specifically limited.

[0060] The above step 103 and its related contents, as an inventive point of an embodiment of the present disclosure, solve the technical problem 2 mentioned in the background technology, "the method based on local features will lead to too high feature dimensions and increase the computational complexity". The factors that lead to increased computational complexity are often as follows: the method based on local features will lead to too high feature dimensions and increase computational complexity. If the above factors are solved, the computational complexity can be greatly reduced, the convergence efficiency of model training can be improved, and the recognition time of the model can be shortened. In order to achieve this effect, first, by setting a three-branch network in the pedestrian re-identification model, different features of pedestrians are recognized respectively. It is convenient to decouple the feature associations at the physical level of the image and the semantic level of the pedestrian. Therefore, it can be used to improve the efficiency of model training. Specifically, by introducing two auxiliary branches, the human body key point branch network and the character attribute branch network, it is convenient to jointly identify and constrain the character identity branch network (i.e., the main network), and realize the decoupling of features at the channel level. Thus, the computational complexity can be greatly reduced. In addition, the convergence efficiency of model training is improved, and the recognition time of the model is shortened.

[0061] Step 104 , performing information association processing on each of the pedestrian identification information in each pedestrian identification information group in the pedestrian identification information group set to generate an associated pedestrian identification information set.

[0062] In some embodiments, the execution entity may perform information association processing on each of the pedestrian identification information in each pedestrian identification information group in the pedestrian identification information group set to generate an associated pedestrian identification information set.

[0063] In some optional implementations of some embodiments, the person identity classification information in each pedestrian recognition information in the pedestrian recognition information group set may include a pedestrian identification. And the execution subject performs information association processing on each pedestrian recognition information in each pedestrian recognition information group in the pedestrian recognition information group set to generate an associated pedestrian recognition information set, which may include the following steps:

[0064] The first step is to perform the following matching steps for the human body key point recognition information included in each pedestrian information in the above pedestrian recognition information group set:

[0065] Sub-step one, in response to determining that the coordinates of each human key point in the human key point coordinate group in the above human key point recognition information are less than the preset number of key points, projecting each human key point coordinate in the above human key point coordinate group to a preset camera coordinate system to obtain a projected key point coordinate group. Among them, the coordinates of each human key point in the human key point coordinate group in the human key point recognition information are less than the preset number of key points, which can represent that the person in the image is blocked and not all key points are recognized. Secondly, the preset camera coordinate system can be the coordinate system of any one of the at least two cameras. Then, the intrinsic parameter matrix and the extrinsic parameter matrix of the camera can be obtained. Finally, the intrinsic parameter matrix and the extrinsic parameter matrix can be used to project the coordinates of each human key point from the image coordinate system to the preset camera coordinate system by coordinate conversion.

[0066] Sub-step two, fill in the coordinates of each post-projection key point in the above-mentioned post-projection key point coordinate group to obtain a filled-in key point coordinate group. Among them, the above-mentioned filled-in key point coordinate group may include a preset number of filled-in key point coordinates. Each filled-in key point coordinate may correspond to different parts of the human body to correspond to different key point labels. Filling is to predict the coordinates of other human key points based on the existing post-projection key point coordinates. Here, the human body three-dimensional geometric structure with the greatest similarity to the coordinates of each post-projection key point in the above-mentioned post-projection key point coordinate group can be matched from a preset human body three-dimensional geometric structure library. Specifically, the human body three-dimensional geometric structure library can be a human body structure model obtained by mapping the two-dimensional space image of the human body to the three-dimensional space in advance. The human body three-dimensional geometric structure library may include human body three-dimensional geometric structures of various actions and body shapes. In addition, each human body three-dimensional geometric structure can also correspond to the key point coordinates of each part. Therefore, the key point coordinates in the corresponding three-dimensional geometric structure of the human body can be matched with the post-projection key point coordinates in the above-mentioned post-projection key point coordinate group, so as to determine the key point coordinates with a matching similarity lower than a preset threshold (i.e., unrecognized human body parts) and the post-projection key point coordinates as the completed key point coordinates to obtain the completed key point coordinate group.

[0067] Sub-step three, determining the above-mentioned padded key point coordinate group and the key point labels corresponding to each padded key point coordinate as the human body key point identification information. Among them, for the key point coordinates in the three-dimensional geometric structure of the human body and the selected padded key point coordinates, the corresponding key point labels can be determined according to the positions of the key point coordinates in the three-dimensional geometric structure of the human body.

[0068] The second step is to match each pedestrian recognition information in the above pedestrian recognition information group set to obtain a matched pedestrian recognition information group set. Wherein, each matched pedestrian recognition information in each matched pedestrian recognition information group in the above matched pedestrian recognition information group set may be: information of the same pedestrian recognized from different pedestrian images. The matching process may be matching pedestrian recognition information including the same pedestrian identifier.

[0069] The third step is to delete the matched pedestrian recognition information group including a matched pedestrian recognition information in the matched pedestrian recognition information group set to obtain a deleted pedestrian recognition information group set. The matched pedestrian recognition information group including a matched pedestrian recognition information can represent pedestrian recognition information that does not match the same pedestrian identifier.

[0070] In practice, for pedestrian identification information that is not matched to the same pedestrian identification, the matched pedestrian identification information can be directly determined as the identification result.

[0071] Step 4: for each deleted pedestrian identification information group in the deleted pedestrian identification information group set, perform the following fusion steps to generate associated pedestrian identification information in the associated pedestrian identification information set:

[0072] Sub-step 1, in response to determining that the personal identity classification information and the personal attribute classification information included in each deleted pedestrian recognition information in the above-mentioned deleted pedestrian recognition information group are the same, the complemented key point coordinate groups included in each deleted pedestrian recognition information in the above-mentioned deleted pedestrian recognition information group are fused to obtain a fused key point coordinate group. Among them, the fusion processing can be to fuse the corresponding complemented key point coordinates in each complemented key point coordinate group. Specifically, the center point coordinates of each complemented key point coordinate corresponding to the same key point label in each complemented key point coordinate group can be determined as the fused key point coordinates to obtain a fused key point coordinate group.

[0073] Sub-step 1: Determine the above-mentioned fused key point coordinate group and any person identity classification information and person attribute classification information included in each deleted pedestrian recognition information as the associated pedestrian recognition information.

[0074] Step 105 , based on the associated pedestrian recognition information set, perform pedestrian re-recognition on the processed pedestrian image to obtain a pedestrian re-recognition result.

[0075] In some embodiments, the execution entity may perform pedestrian re-identification on the processed pedestrian image based on the associated pedestrian identification information set to obtain a pedestrian re-identification result.

[0076] In some optional implementations of some embodiments, the execution subject performs pedestrian re-identification on the processed pedestrian image based on the associated pedestrian recognition information set to obtain a pedestrian re-identification result, which may include the following steps:

[0077] The first step is to project each fused key point coordinate in the fused key point coordinate group included in the above-mentioned associated pedestrian recognition information set to the corresponding processed pedestrian image, and associate the projected key point coordinates to obtain the projected pedestrian image. The association may be to connect each key point coordinate corresponding to each pedestrian in a preset human body part order. Secondly, the projection may be to project the fused key point coordinates from the camera coordinate system to the corresponding processed pedestrian image by coordinate transformation using the above-mentioned intrinsic parameter matrix and extrinsic parameter matrix.

[0078] In practice, the key points in the image may be 1-4 pixels in size, and the connecting lines may be only one or two pixels wide. Therefore, the key point features of each pedestrian can be increased without affecting the extraction of image features.

[0079] In the second step, the projected pedestrian image is input again into the pedestrian re-identification model to generate a re-identification information group. Each re-identification information in the re-identification information group corresponds to a pedestrian. Here, after the pedestrian re-identification model is trained, the human key point branch network and the character attribute branch network in the model can be deleted. Only the output of the character identity branch network in the pedestrian re-identification model is used as the re-identification information group. In this way, the model can only output the pedestrian identity information to achieve pedestrian re-identification of different images. Thereby, the amount of calculation of the model in the application process can be reduced and the recognition efficiency can be improved. The output of the three branches in the pedestrian re-identification model can also be used as the re-identification information group. Thereby, the accuracy of the model output results is improved on the basis of pedestrian re-identification.

[0080] The third step is to determine each generated pedestrian re-identification information group as a pedestrian re-identification result.

[0081] The above steps 104-105 and their related contents are an inventive point of the embodiment of the present disclosure, which solves the technical problem 3 mentioned in the background technology: "If the person in the image is occluded, the missing body parts will affect the corresponding feature channels, making it difficult for the model training to converge, thereby affecting the training of other channels, and thus making it difficult for the trained model to better identify the occluded person." The factors that make it difficult for the trained model to better identify the occluded person are often as follows: If the person in the image is occluded, the missing body parts will affect the corresponding feature channels, making it difficult for the model training to converge, thereby affecting the training of other channels. If the above factors are solved, the recognition ability of the occluded pedestrians in the image can be improved. In order to achieve this effect, firstly, by projection, the coordinates of the key points of the pedestrians taken from different perspectives can be projected to the same camera coordinate system to facilitate the association of multi-perspective features. Also, because the features of the images under multiple perspectives are combined, the features can be expanded to facilitate feature extraction. Then, considering that the features of the occluded pedestrians are difficult to identify and determine, the features of the occluded area can be added to the image by filling in the key point coordinates. This makes the features of each pedestrian more complete and reduces the impact on the corresponding feature channels. This facilitates the convergence efficiency of model training. Furthermore, it can also improve the recognition ability of occluded pedestrians.

[0082] The above-mentioned embodiments of the present disclosure have the following beneficial effects: through the pedestrian re-identification method of some embodiments of the present disclosure, the robustness of identifying occlusion and posture changes in pedestrian images can be improved. Specifically, the reason for the less robustness to occlusion and posture changes is that the global feature method directly extracts global features from the entire person image. Based on this, the pedestrian re-identification method of some embodiments of the present disclosure first obtains pedestrian images taken by at least two cameras to obtain a pedestrian image group. Among them, each pedestrian image in the above pedestrian image group is taken by different cameras at the same time. Then, each pedestrian image in the above pedestrian image group is image preprocessed to obtain a processed pedestrian image group. After that, pedestrian recognition is performed on each processed pedestrian image in the above processed pedestrian image group to generate a pedestrian recognition information group set. Among them, each pedestrian recognition information group in the above pedestrian recognition information group set corresponds to a pedestrian, and each pedestrian recognition information in each pedestrian recognition information group is obtained by identifying different processed pedestrian images. Here, pedestrian recognition can be used to preliminarily detect pedestrian information in the image. Next, information association processing is performed on each of the pedestrian recognition information in each pedestrian recognition information group in the above-mentioned pedestrian recognition information group set to generate an associated pedestrian recognition information set. Here, through the association processing, the features in multiple images can be better combined to highlight the features of occluded people and people with posture changes in the image. This facilitates feature extraction. Finally, based on the above-mentioned associated pedestrian recognition information set, pedestrian re-identification is performed on the above-mentioned processed pedestrian image to obtain a pedestrian re-identification result. Therefore, through pedestrian re-identification, the pedestrian information in the image can be re-identified on the basis of facilitating the extraction of the features of occluded people and people with posture changes in the image. Furthermore, the robustness of the recognition of occlusion and posture changes in pedestrian images can be improved.

[0083] Further references Figure 2 As an implementation of the methods shown in the above figures, the present disclosure provides some embodiments of a pedestrian re-identification device. These device embodiments are similar to Figure 1 Corresponding to the method embodiments shown, the device can be specifically applied to various electronic devices.

[0084] like Figure 2As shown, the pedestrian re-identification device 200 of some embodiments includes: an acquisition unit 201, a preprocessing unit 202, a first recognition unit 203, an association processing unit 204 and a second recognition unit 205. The acquisition unit 201 is configured to acquire pedestrian images taken by at least two cameras to obtain a pedestrian image group, wherein each pedestrian image in the pedestrian image group is taken by different cameras at the same time; the preprocessing unit 202 is configured to perform image preprocessing on each pedestrian image in the pedestrian image group to obtain a processed pedestrian image group; the first recognition unit 203 is configured to perform pedestrian recognition on each processed pedestrian image in the processed pedestrian image group to generate a pedestrian recognition information group set, wherein the pedestrian recognition information Each pedestrian identification information group in the group set corresponds to a pedestrian, and each pedestrian identification information in each pedestrian identification information group is identified from different processed pedestrian images; the association processing unit 204 is configured to perform information association processing on each pedestrian identification information in each pedestrian identification information group in the above-mentioned pedestrian identification information group set to generate an associated pedestrian identification information set; the second identification unit 205 is configured to perform pedestrian re-identification on the above-mentioned processed pedestrian image based on the above-mentioned associated pedestrian identification information set to obtain a pedestrian re-identification result.

[0085] It is understood that the units described in the device 200 are similar to those described in the reference Figure 1 Therefore, the operations, features and beneficial effects described above for the method are also applicable to the device 200 and the units included therein, and will not be described in detail here.

[0086] Reference below Figure 3 , which shows a structural schematic diagram of an electronic device 300 suitable for implementing some embodiments of the present disclosure. Figure 3 The electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0087] like Figure 3 As shown, the electronic device 300 may include a processing device 301 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 302 or a program loaded from a storage device 308 into a random access memory (RAM) 303. In the RAM 303, various programs and data required for the operation of the electronic device 300 are also stored. The processing device 301, the ROM 302, and the RAM 303 are connected to each other via a bus 304. An input / output (I / O) interface 305 is also connected to the bus 304.

[0088] Typically, the following devices may be connected to the I / O interface 305: input devices 306 including, for example, a touch screen, a touch pad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; output devices 307 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; storage devices 308 including, for example, a magnetic tape, a hard disk, etc.; and communication devices 309. The communication devices 309 may allow the electronic device 300 to communicate with other devices wirelessly or by wire to exchange data. Although Figure 3 The electronic device 300 is shown with various devices, but it should be understood that it is not required to implement or possess all the devices shown. More or fewer devices may be implemented or possessed instead. Figure 3 Each block shown in the figure may represent one device, or may represent multiple devices as required.

[0089] In particular, according to some embodiments of the present disclosure, the process described above with reference to the flowchart can be implemented as a computer software program. For example, some embodiments of the present disclosure include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In some such embodiments, the computer program can be downloaded and installed from the network through the communication device 309, or installed from the storage device 308, or installed from the ROM 302. When the computer program is executed by the processing device 301, the above-mentioned functions defined in the method of some embodiments of the present disclosure are executed.

[0090] It should be noted that the computer-readable medium in some embodiments of the present disclosure may be a computer-readable signal medium or a computer-readable storage medium or any combination of the two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, 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 wires, 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 some embodiments of the present disclosure, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, device or device. In some embodiments of the present disclosure, the 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. This propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. The computer readable signal medium may also be any computer readable 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, apparatus or device. The program code contained on the computer readable medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.

[0091] In some embodiments, the client and the server may communicate using any currently known or future developed network protocol such as HTTP (Hyper Text Transfer Protocol), and may be interconnected with any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network ("LAN"), a wide area network ("WAN"), an internet (e.g., the Internet), and a peer-to-peer network (e.g., an ad hoc peer-to-peer network), as well as any currently known or future developed network.

[0092] The computer-readable medium may be included in the device, or may exist independently without being incorporated into the electronic device. The computer-readable medium carries one or more programs. When the one or more programs are executed by the electronic device, the electronic device: obtains pedestrian images taken by at least two cameras to obtain a pedestrian image group, wherein each pedestrian image in the pedestrian image group is taken by different cameras at the same time; performs image preprocessing on each pedestrian image in the pedestrian image group to obtain a processed pedestrian image group; performs pedestrian recognition on each processed pedestrian image in the processed pedestrian image group to generate a pedestrian recognition information group set, wherein each pedestrian recognition information group in the pedestrian recognition information group set corresponds to a pedestrian, and each pedestrian recognition information in each pedestrian recognition information group is obtained by recognition from different processed pedestrian images; performs information association processing on each pedestrian recognition information in each pedestrian recognition information group in the pedestrian recognition information group set to generate an associated pedestrian recognition information set; based on the associated pedestrian recognition information set, performs pedestrian re-identification on the processed pedestrian images to obtain a pedestrian re-identification result.

[0093] Computer program code for performing the operations of some embodiments of the present disclosure 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 cases involving 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).

[0094] 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 disclosure. 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 implementations as replacements, 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 the specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0095] The units described in some embodiments of the present disclosure may be implemented by software or hardware. The units described may also be provided in a processor, for example, may be described as: a processor comprising: an acquisition unit, a preprocessing unit, a first recognition unit, an association processing unit, and a second recognition unit. The names of these units do not, in some cases, constitute limitations on the units themselves, for example, the acquisition unit may also be described as a "unit for acquiring pedestrian images".

[0096] The functions described above herein may be performed at least in part by one or more hardware logic components. For example, without limitation, exemplary types of hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), and the like.

[0097] The above descriptions are only some preferred embodiments of the present disclosure and an explanation of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the embodiments of the present disclosure is not limited to the technical solutions formed by a specific combination of the above-mentioned technical features, but should also cover other technical solutions formed by any combination of the above-mentioned technical features or their equivalent features without departing from the above-mentioned inventive concept. For example, the above-mentioned features are replaced with the technical features with similar functions disclosed in the embodiments of the present disclosure (but not limited to) and the technical solutions formed.

Claims

1. A pedestrian re-identification method, comprising: Acquire pedestrian images taken by at least two cameras to obtain a pedestrian image group, wherein each pedestrian image in the pedestrian image group is taken by different cameras at the same time; Performing image preprocessing on each pedestrian image in the pedestrian image group to obtain a processed pedestrian image group; Performing pedestrian recognition on each processed pedestrian image in the processed pedestrian image group to generate a pedestrian recognition information group set, wherein each pedestrian recognition information group in the pedestrian recognition information group set corresponds to a pedestrian, and each pedestrian recognition information in each pedestrian recognition information group is obtained by recognition from different processed pedestrian images; Performing information association processing on each of the pedestrian identification information in each pedestrian identification information group in the pedestrian identification information group set to generate an associated pedestrian identification information set; Based on the associated pedestrian recognition information set, performing pedestrian re-recognition on the processed pedestrian image to obtain a pedestrian re-recognition result; The person identity classification information in each pedestrian identification information in the pedestrian identification information group set includes a pedestrian identification; and The step of performing information association processing on each of the pedestrian identification information in each pedestrian identification information group in the pedestrian identification information group set to generate an associated pedestrian identification information set includes: For the human body key point recognition information included in each pedestrian information in the pedestrian recognition information group set, the following matching steps are performed: In response to determining that the coordinates of each human body key point in the human body key point coordinate group in the human body key point recognition information are less than the preset number of key points, projecting the coordinates of each human body key point in the human body key point coordinate group to a preset camera coordinate system to obtain a projected key point coordinate group; The post-projection key point coordinates in the post-projection key point coordinate group are padded to obtain a padded key point coordinate group, wherein the padded key point coordinate group includes a preset number of padded key point coordinates, each padded key point coordinate corresponds to a different part of the human body to correspond to a different key point label, and the padded key point coordinates are used to predict the coordinates of other human body key points based on the existing post-projection key point coordinates; Determine the key point coordinate group after completion and the key point labels corresponding to each key point coordinate after completion as human body key point recognition information; Performing matching processing on each pedestrian identification information in the pedestrian identification information group set to obtain a matched pedestrian identification information group set, wherein each matched pedestrian identification information in each matched pedestrian identification information group in the matched pedestrian identification information group set is: information of the same pedestrian identified from different pedestrian images, and the matching processing is matching pedestrian identification information including the same pedestrian identification; Deleting the matched pedestrian recognition information group including one matched pedestrian recognition information in the matched pedestrian recognition information group set to obtain a deleted pedestrian recognition information group set, wherein the matched pedestrian recognition information group including one matched pedestrian recognition information represents pedestrian recognition information that is not matched to the same pedestrian identifier; For each deleted pedestrian identification information group in the deleted pedestrian identification information group set, the following fusion steps are performed to generate associated pedestrian identification information in the associated pedestrian identification information set: In response to determining that the person identity classification information and the person attribute classification information included in each deleted pedestrian recognition information in the deleted pedestrian recognition information group are the same, fusing the complemented key point coordinate groups included in each deleted pedestrian recognition information in the deleted pedestrian recognition information group to obtain a fused key point coordinate group, wherein the fusion processing is to fuse the corresponding complemented key point coordinates in each complemented key point coordinate group; The fused key point coordinate group and any person identity classification information and person attribute classification information included in each deleted pedestrian recognition information are determined as associated pedestrian recognition information.

2. The method according to claim 1, wherein: The method further comprises: Mapping the pedestrian re-identification result to each pedestrian image in the pedestrian image group to obtain a mapped pedestrian image group, wherein each mapped pedestrian image in the mapped pedestrian image group has marked the same pedestrian; The mapped pedestrian image group is sent to a display terminal for display.

3. The method according to claim 1, wherein: The performing image preprocessing on each pedestrian image in the pedestrian image group to obtain a processed pedestrian image group includes: Performing rough person detection on each pedestrian image in the pedestrian image group to obtain a rough person detection information group, wherein each rough person detection information in the rough person detection information group indicates whether there is a pedestrian in the pedestrian image; In response to determining that each piece of rough person detection information in the rough person detection information group is information indicating the presence of a pedestrian, graying each pedestrian image in the pedestrian image group to generate a pedestrian grayscale image group; Performing contrast adjustment on each pedestrian grayscale image in the pedestrian grayscale image group to obtain an adjusted pedestrian image group; Based on the rough person detection information group, grayscale slicing is performed on each adjusted pedestrian image in the adjusted pedestrian image group to obtain a processed pedestrian image group, wherein the grayscale slicing distinguishes the background area and the pedestrian area in the pedestrian image for preliminary removal of the background area.

4. The method according to claim 1, wherein: The performing pedestrian recognition on each processed pedestrian image in the processed pedestrian image group to generate a pedestrian recognition information group set includes: For each processed pedestrian image in the processed pedestrian image group, the following recognition steps are performed to generate a pedestrian recognition information group in the pedestrian recognition information group set: Input the processed pedestrian image into the backbone network of a preset pedestrian re-identification model to extract features, and obtain an initial feature map of the image, wherein the pedestrian re-identification model also includes a branch network module, and the branch network module includes: a person identity branch network, a human body key point branch network, and a person attribute branch network; Inputting the initial feature map of the image into the person identity branch network to output person identity classification information corresponding to each pedestrian in the processed pedestrian image; Inputting the initial feature map of the image into the human key point branch network to output a set of human key point heat map groups corresponding to each pedestrian in the processed pedestrian image; Determine a human key point recognition information group corresponding to the human key point heat map group set, wherein each human key point recognition information in the human key point recognition information group includes a human key point coordinate group and a corresponding key point label group; Inputting the image initial feature map into the character attribute branch network to output character attribute classification information corresponding to each pedestrian in the processed pedestrian image; The person identity classification information, the human body key point recognition information group and the person attribute classification information corresponding to each pedestrian in the processed pedestrian image are determined as pedestrian recognition information to obtain a pedestrian recognition information group.

5. The method according to claim 4, wherein: The pedestrian re-identification model is trained and generated by the following steps: Acquire a sample image and a corresponding sample label, wherein the sample image includes at least one pedestrian, and the sample label includes a pedestrian label group corresponding to each pedestrian; Inputting the sample image into an initial backbone network of an initial person re-identification model to obtain an extracted person feature map, wherein the initial person re-identification model further includes: an initial person identity branch network, an initial human body key point branch network, and an initial person attribute branch network; The extracted pedestrian feature map is respectively input into the initial person identity branch network, the initial human body key point branch network and the initial person attribute branch network to obtain a person identity classification identification group, a human body key point heat map group set and a person attribute classification identification group set, wherein the human body key point heat map in the human body key point heat map group set is a heat map centered on the human body key point, and each person identity classification identification, human body key point heat map group and person attribute classification identification group corresponds to a pedestrian; Determine a person identity loss value between each person identity classification identifier in the person identity classification identifier group and the pedestrian label group; Determine a human key point loss value between the human key point heat map group set and the pedestrian label group; Determine a person attribute loss value between the person attribute classification identification group set and the pedestrian label group; Determine the total loss value as a weighted sum of the character identity loss value, the character key point loss value, and the character attribute loss value; In response to determining that the total loss value is less than or equal to a preset loss threshold, and the training of the sample image meets a preset training end condition, the initial pedestrian re-identification model is determined as a trained pedestrian re-identification model.

6. The method according to claim 1, wherein: The step of performing pedestrian re-identification on the processed pedestrian image based on the associated pedestrian identification information set to obtain a pedestrian re-identification result includes: Projecting each fused key point coordinate in the fused key point coordinate group included in the associated pedestrian recognition information in the associated pedestrian recognition information set to the corresponding processed pedestrian image, and associating the projected key point coordinates to obtain a projected pedestrian image, wherein the association is to connect each key point coordinate corresponding to each pedestrian according to a preset human body part order; Inputting the projected pedestrian image into the pedestrian re-identification model again to generate a re-identification information group, wherein each re-identification information in the re-identification information group corresponds to a pedestrian; The generated pedestrian re-identification information groups are determined as pedestrian re-identification results.

7. A pedestrian re-identification device, comprising: an acquisition unit, configured to acquire pedestrian images taken by at least two cameras to obtain a pedestrian image group, wherein each pedestrian image in the pedestrian image group is taken by different cameras at the same time; a preprocessing unit configured to perform image preprocessing on each pedestrian image in the pedestrian image group to obtain a processed pedestrian image group; a first recognition unit configured to perform pedestrian recognition on each processed pedestrian image in the processed pedestrian image group to generate a pedestrian recognition information group set, wherein each pedestrian recognition information group in the pedestrian recognition information group set corresponds to a pedestrian, and each pedestrian recognition information in each pedestrian recognition information group is obtained by recognition from different processed pedestrian images; an association processing unit configured to perform information association processing on each of the pedestrian identification information in each pedestrian identification information group in the pedestrian identification information group set to generate an associated pedestrian identification information set; A second recognition unit is configured to perform pedestrian re-recognition on the processed pedestrian image based on the associated pedestrian recognition information set to obtain a pedestrian re-recognition result; The person identity classification information in each pedestrian identification information in the pedestrian identification information group set includes a pedestrian identification; and The step of performing information association processing on each of the pedestrian identification information in each pedestrian identification information group in the pedestrian identification information group set to generate an associated pedestrian identification information set includes: For the human body key point recognition information included in each pedestrian information in the pedestrian recognition information group set, the following matching steps are performed: In response to determining that the coordinates of each human body key point in the human body key point coordinate group in the human body key point recognition information are less than the preset number of key points, projecting the coordinates of each human body key point in the human body key point coordinate group to a preset camera coordinate system to obtain a projected key point coordinate group; The post-projection key point coordinates in the post-projection key point coordinate group are padded to obtain a padded key point coordinate group, wherein the padded key point coordinate group includes a preset number of padded key point coordinates, each padded key point coordinate corresponds to a different part of the human body to correspond to a different key point label, and the padded key point coordinates are used to predict the coordinates of other human body key points based on the existing post-projection key point coordinates; Determine the key point coordinate group after completion and the key point labels corresponding to each key point coordinate after completion as human body key point recognition information; Performing matching processing on each pedestrian identification information in the pedestrian identification information group set to obtain a matched pedestrian identification information group set, wherein each matched pedestrian identification information in each matched pedestrian identification information group in the matched pedestrian identification information group set is: information of the same pedestrian identified from different pedestrian images, and the matching processing is matching pedestrian identification information including the same pedestrian identification; Deleting the matched pedestrian recognition information group including one matched pedestrian recognition information in the matched pedestrian recognition information group set to obtain a deleted pedestrian recognition information group set, wherein the matched pedestrian recognition information group including one matched pedestrian recognition information represents pedestrian recognition information that is not matched to the same pedestrian identifier; For each deleted pedestrian identification information group in the deleted pedestrian identification information group set, the following fusion steps are performed to generate associated pedestrian identification information in the associated pedestrian identification information set: In response to determining that the person identity classification information and the person attribute classification information included in each deleted pedestrian recognition information in the deleted pedestrian recognition information group are the same, fusing the complemented key point coordinate groups included in each deleted pedestrian recognition information in the deleted pedestrian recognition information group to obtain a fused key point coordinate group, wherein the fusion processing is to fuse the corresponding complemented key point coordinates in each complemented key point coordinate group; The fused key point coordinate group and any person identity classification information and person attribute classification information included in each deleted pedestrian recognition information are determined as associated pedestrian recognition information.

8. An electronic device comprising: one or more processors; a storage device having one or more programs stored thereon, When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 6.

9. A computer readable medium having a computer program stored thereon, wherein: When the computer program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images