CN114783037A - Object re-recognition method, object re-recognition apparatus, and computer-readable storage medium - Google Patents

Object re-recognition method, object re-recognition apparatus, and computer-readable storage medium Download PDF

Info

Publication number
CN114783037A
CN114783037A CN202210689509.6A CN202210689509A CN114783037A CN 114783037 A CN114783037 A CN 114783037A CN 202210689509 A CN202210689509 A CN 202210689509A CN 114783037 A CN114783037 A CN 114783037A
Authority
CN
China
Prior art keywords
target
human body
image
features
snapshot
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202210689509.6A
Other languages
Chinese (zh)
Other versions
CN114783037B (en
Inventor
唐邦杰
潘华东
殷俊
金恒
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Zhejiang Dahua Technology Co Ltd
Original Assignee
Zhejiang Dahua Technology Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Zhejiang Dahua Technology Co Ltd filed Critical Zhejiang Dahua Technology Co Ltd
Priority to CN202210689509.6A priority Critical patent/CN114783037B/en
Publication of CN114783037A publication Critical patent/CN114783037A/en
Application granted granted Critical
Publication of CN114783037B publication Critical patent/CN114783037B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/22—Matching criteria, e.g. proximity measures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/25—Fusion techniques
    • G06F18/253—Fusion techniques of extracted features

Abstract

The application discloses a target re-identification method, a target re-identification device and a computer readable storage medium, wherein the method comprises the following steps: processing a video to be detected to obtain a snapshot image set aiming at a snapshot target; responding to a first image and second images corresponding to a plurality of postures of a snapshot target in a snapshot image set, processing a human face contained in the first image to obtain human face characteristics of the snapshot target, and processing human bodies contained in the second images corresponding to the plurality of postures to obtain head and shoulder characteristics of the plurality of postures and human body characteristics of the plurality of postures; fusing the head-shoulder characteristics of a plurality of postures to obtain multi-posture head-shoulder characteristics, and fusing the human body characteristics of a plurality of postures to obtain multi-posture human body characteristics; and generating a re-recognition result based on the human face features, the multi-pose human body features and the multi-pose head-shoulder features, wherein the re-recognition result comprises whether the snapshot target is a historical snapshot target. Through the mode, the de-weight accuracy rate can be improved.

Description

Object re-recognition method, object re-recognition apparatus, and computer-readable storage medium
Technical Field
The present application relates to the field of video analysis technologies, and in particular, to a target re-identification method, a target re-identification apparatus, and a computer-readable storage medium.
Background
The requirement of personnel identification exists in the application fields of intelligent police, intelligent retail, intelligent real estate or intelligent scenic spots and the like, and the personnel identification can be used for not only falling to the ground but also accurately counting the passenger flow. In practical application, certain application scenes do not need to make personnel identity explicit, and whether repeated personnel appear is judged through re-identification (ReID) so as to provide more accurate statistical data for industrial application, but the accuracy of a re-identification scheme is not high, so that the subsequent processing effect is poor.
Disclosure of Invention
The application provides a target re-identification method, a target re-identification device and a computer readable storage medium, which can improve the de-duplication accuracy.
In order to solve the technical problem, the technical scheme adopted by the application is as follows: a target re-identification method is provided, and the method comprises the following steps: processing a video to be detected to obtain a snapshot image set aiming at a snapshot target; responding to a first image and second images corresponding to a plurality of postures of a snapshot target in a snapshot image set, processing a human face contained in the first image to obtain human face characteristics of the snapshot target, and processing human bodies contained in the second images corresponding to the plurality of postures to obtain head and shoulder characteristics of the plurality of postures and human body characteristics of the plurality of postures; fusing the head-shoulder characteristics of a plurality of postures to obtain multi-posture head-shoulder characteristics, and fusing the human body characteristics of a plurality of postures to obtain multi-posture human body characteristics; and generating a re-recognition result based on the human face features, the multi-pose human body features and the multi-pose head-shoulder features, wherein the re-recognition result comprises whether the snapshot target is a historical snapshot target.
In order to solve the above technical problem, another technical solution adopted by the present application is: there is provided an object re-recognition apparatus comprising a memory and a processor connected to each other, wherein the memory is used for storing a computer program, and the computer program, when executed by the processor, is used for implementing the object re-recognition method in the above technical solution.
In order to solve the above technical problem, another technical solution adopted by the present application is: there is provided a computer readable storage medium for storing a computer program for implementing the object re-identification method in the above technical solution when the computer program is executed by a processor.
Through above-mentioned scheme, this application's beneficial effect is: firstly, processing a to-be-detected video acquired in real time to obtain a snapshot image set of a snapshot target; if the first image and the second images corresponding to the plurality of postures of the snapshot target exist in the snapshot image set, processing the human face contained in the first image to obtain the human face characteristics of the snapshot target, and processing the human body contained in each second image to obtain the head-shoulder characteristics and the human body characteristics; fusing the head-shoulder characteristics of all postures to obtain multi-posture head-shoulder characteristics, and fusing the human body characteristics of all postures to obtain multi-posture human body characteristics; generating a re-recognition result of whether the snapshot target is a historical snapshot target or not based on the human face features, the multi-pose human body features and the multi-pose head and shoulder features; according to the scheme, the human body features and the head and shoulder features of the snapshot target in different postures are extracted in real time and are fused, the human body features and the head and shoulder features are combined with the human face features, whether the snapshot target is a repeated target or not is judged, and due to the fact that the types of the features are richer, more features can be referred to during feature comparison, the problem that the error rate is high due to the fact that single features are used for comparison can be solved, and the duplicate removal accuracy rate is effectively improved; moreover, due to the fact that the features under different postures are fused, the method has good robustness in the conditions that the face is shielded or the human body is not over against the camera device, and the like, and the application range of the scheme is greatly enlarged.
Drawings
In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the description of the embodiments are briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present application, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without creative efforts. Wherein:
FIG. 1 is a schematic flow chart diagram illustrating an embodiment of a target re-identification method provided in the present application;
FIG. 2 is a schematic flow chart diagram illustrating another embodiment of a method for object re-identification provided in the present application;
FIG. 3 is a schematic diagram illustrating an embodiment of a target re-identification apparatus provided in the present application;
FIG. 4 is a schematic structural diagram of another embodiment of the object re-identification device provided in the present application;
FIG. 5 is a schematic structural diagram of a target snapshot model in the embodiment shown in FIG. 4;
FIG. 6 is a schematic structural diagram of an embodiment of a computer-readable storage medium provided in the present application.
Detailed Description
The present application will be described in further detail with reference to the following drawings and examples. It is to be noted that the following examples are only illustrative of the present application, and do not limit the scope of the present application. Likewise, the following examples are only some examples of the present application, not all examples, and all other examples obtained by a person of ordinary skill in the art without making any creative effort fall within the protection scope of the present application.
Reference in the specification to "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment can be included in at least one embodiment of the specification. The appearances of the phrase in various places in the specification are not necessarily all referring to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments. It is explicitly and implicitly understood by one skilled in the art that the embodiments described herein can be combined with other embodiments.
It should be noted that the terms "first", "second" and "third" in the present application are used for descriptive purposes only and are not to be construed as indicating or implying relative importance or implying any number of indicated technical features. Thus, a feature defined as "first," "second," or "third" may explicitly or implicitly include at least one of the feature. In the description of the present application, "plurality" means at least two, e.g., two, three, etc., unless explicitly specified otherwise. Furthermore, the terms "include" and "have," as well as any variations thereof, are intended to cover a non-exclusive inclusion. For example, a process, method, system, article, or apparatus that comprises a list of steps or elements is not limited to only those steps or elements listed, but may alternatively include other steps or elements not listed, or inherent to such process, method, article, or apparatus.
Terms of art related to the present application will be described.
Target detection: and positioning the targets appearing in the input images based on the pre-trained model.
Image re-identification: extracting the characteristic vectors of the human body image or the human body part image (the image of a certain part of the human body), and calculating the similarity of the characteristic vectors of different people through similarity measurement modes such as cosine distance and the like, thereby judging whether the people are the same.
Whether the personnel are the same or not is judged through face feature similarity comparison in the correlation technique, the purpose of removing the weight of the personnel is achieved, however, due to the complexity of installation arrangement and an actual scene, the probability of appearance of a high-quality face is low, when the distance between the personnel and the camera equipment is long, the personnel face back to the camera equipment (namely the posture of the personnel is the back posture), the personnel face side to the camera equipment (namely the posture of the personnel is the side posture) or the face of the personnel is shielded, the effect is poor, the personnel removing weight and the personnel identification only depend on the face features, and therefore, how to effectively identify the identity of the personnel and remove the weight is important. Based on this, the application provides a new scheme, combines people's face characteristic, human body characteristic and head and shoulder characteristic, realizes heavy discernment, can promote heavy discernment's rate of accuracy, and the real-time is better, describes in detail below.
Referring to fig. 1, fig. 1 is a schematic flowchart of an embodiment of a target re-identification method provided in the present application, an execution subject of the embodiment is a target re-identification device, the target re-identification device may be an image capturing apparatus or a computer, and the like, which has processing capabilities, and the method includes:
s11: and processing the video to be detected to obtain a snapshot image set aiming at the snapshot target.
The method comprises the steps that a video to be detected can be obtained from a database, the video to be detected comprises a snapshot target, and the snapshot target can be a person; or shooting a target scene (such as a shopping mall, a scenic spot or a station) in real time by adopting the camera equipment to obtain a real-time video to be detected; or receiving the video to be detected sent by the front-end equipment in real time.
After the video to be detected is obtained, the video to be detected is processed (such as detection, tracking or identification) by adopting a video processing scheme in the related technology, so that a snapshot image set related to a snapshot target is obtained. Specifically, the snapshot targets correspond to the snapshot image set in a one-to-one mode, the snapshot image set comprises second images corresponding to a plurality of postures of the snapshot targets, the second images contain human bodies, and the postures comprise forward postures, lateral postures or backward postures. Alternatively, the snap shot image set includes second images of the first image corresponding to a plurality of poses of the snap shot target, the first image contains a human face, the first image is different from all the second images, or the first image may be a part of one of the second images corresponding to the plurality of poses, for example, the second images are marked as I1 to I4, and the first image may be an image of the second image I2 where the face of the person is located.
S12: responding to the fact that the first image and the second images corresponding to the multiple postures of the snapshot target exist in the snapshot image set, processing the face contained in the first image to obtain the face characteristics of the snapshot target, and processing the human body contained in the second images corresponding to the multiple postures to obtain the head and shoulder characteristics of the multiple postures and the human body characteristics of the multiple postures.
After the snapshot image set is obtained, judging whether a first image and a second image exist in the snapshot image set at the same time; if the first image and the second image exist in the snapshot image set at the same time, processing the image of the face in the first image by adopting a face feature extraction scheme commonly used in the related technology to obtain the face feature of the snapshot target; and processing the image of the human body in the second image by adopting a human body feature extraction scheme commonly used in the related technology to obtain the human body features of the snapshot target.
In a specific embodiment, the second image includes a human body image and a head-shoulder image, the head-shoulder image is a part of the human body image, and the human body image is subjected to feature extraction processing to obtain human body features; and performing feature extraction processing on the head and shoulder images to obtain head and shoulder features.
S13: and fusing the head-shoulder characteristics of the plurality of postures to obtain multi-posture head-shoulder characteristics, and fusing the human body characteristics of the plurality of postures to obtain multi-posture human body characteristics.
Under the condition of dense crowds or shielding of background objects, the human body is possibly shielded to cause incompleteness of the human body, and influence is caused on subsequent characteristic comparison; in order to solve the problem, after the head and shoulder features of different postures are obtained, a feature fusion method in the related technology is adopted to perform fusion processing on the head and shoulder features to obtain multi-posture head and shoulder features; for example: the head and shoulder characteristics of all postures are averaged to obtain multi-posture head and shoulder characteristics, so that the probability of shielding the head and shoulder is lower, and the characteristics are more stable.
In order to fuse the characteristic differences of different angles and postures of the human body, the characteristic descriptor is more robust; after the human body features of different postures are obtained, a feature fusion method in the related technology is adopted to perform fusion processing on the human body features to obtain multi-posture human body features; for example: and averaging the human body characteristics of all the postures to obtain the multi-posture human body characteristics.
It will be appreciated that in addition to the use of the head and shoulder features, features of other parts of the human body may be used, such as: a hand feature or a foot feature. When the snapshot image set only contains second images of certain postures, only human body features corresponding to the second images are fused, and human shoulder features corresponding to the second images are fused. For example, taking the posture as a forward posture, a lateral posture or a backward posture as an example, first, whole-body images (i.e., second images) of the front, the side and the back of the same snapshot target are acquired and recorded as a front whole-body image, a side whole-body image and a back whole-body image; then extracting the human body characteristics of the front whole body image, the side whole body image and the back whole body image, and averaging the human body characteristics to obtain multi-posture human body characteristics; when the snapshot target only appears on the front side, the side face and a part of the back side in the whole process, the human body characteristics of the existing second image are only averaged.
S14: and generating a re-recognition result based on the human face features, the multi-pose human body features and the multi-pose head-shoulder features.
After the relevant features (including the human face features, the multi-pose human body features and the multi-pose head and shoulder features) of the snapshot target are obtained, whether the snapshot target is the same as the historical snapshot target in the target database or not can be judged by comparing the relevant features with the features in the target database established in advance, and a re-recognition result is obtained, wherein the re-recognition result includes whether the snapshot target is the historical snapshot target or not. It can be understood that when the target database is empty, the snapshot target is a new target because there is no comparison object, and the corresponding relevant features are stored in the target database for subsequent use.
Further, thresholds for feature comparison may be set, such as: judging whether the similarity between the face features of the snapshot target and the face features of the historical snapshot target is greater than the first similarity or not, and if so, determining that the face features of the snapshot target and the historical snapshot target are the same; and/or judging whether the similarity between the human body characteristics of the snapshot target and the human body characteristics of the historical snapshot target is greater than a second similarity, if so, determining that the human body characteristics of the snapshot target and the historical snapshot target are the same; and/or judging whether the similarity between the head and shoulder characteristics of the snapshot target and the head and shoulder characteristics of the historical snapshot target is greater than a third similarity, and if so, determining that the head and shoulder characteristics are the same. Alternatively, priority of feature comparison may also be set, such as: setting the priority of the human face features as a first priority, setting the priority of the multi-pose human body features as a second priority, setting the priority of the multi-pose head-shoulder features as a third priority, wherein the first priority is higher than the second priority, and the second priority is higher than the third priority, namely when the features are compared, the similarity (recorded as the human face similarity) between the human face features of the snapshot target and the human face features of the historical snapshot target is compared, and if the human face similarity does not meet the requirement, the multi-pose human body features of the snapshot target and the multi-pose human body features (human body similarity) of the historical snapshot target are compared; if the human body similarity does not meet the requirement, comparing the multi-pose head-shoulder characteristics of the snapshot target with the similarity (recorded as the head-shoulder similarity) of the multi-pose head-shoulder characteristics of the historical snapshot target; and if the head-shoulder similarity does not meet the requirement, the snapshot target is proved to be a new target.
In other embodiments, in response to that the first image does not exist in the snapshot image set but second images corresponding to a plurality of postures of the snapshot target exist, processing human bodies contained in the second images corresponding to the plurality of postures to obtain head and shoulder features of the plurality of postures and human body features of the plurality of postures; and generating a re-recognition result based on the multi-pose human body characteristics and the multi-pose head-shoulder characteristics.
Further, an attribute identification scheme in the related technology can be adopted to perform attribute identification processing on the snap-shot target in the second image to obtain attribute characteristics, wherein the attribute characteristics are used for representing attribute information of the snap-shot target, the attribute information comprises clothes color, age, whether a hat is worn, whether glasses or sunglasses are worn, hair color, clothes texture or gender, and the clothes color comprises coat color or coat-off color; then, based on the multi-pose human body features, the multi-pose head-shoulder features and the attribute features, a re-recognition result is generated.
The embodiment provides a real-time personnel duplication elimination judging method, which includes the steps of extracting features of a second image of the same snapshot target in a video to be detected under multiple postures to obtain multiple human body features and multiple head and shoulder features; respectively performing multi-feature fusion on the plurality of human body features and the plurality of head and shoulder features to obtain multi-pose human body features and multi-pose head and shoulder features; then, carrying out real-time personnel duplication elimination judgment by utilizing the human face characteristics, the multi-pose human body characteristics and the multi-pose head and shoulder characteristics; because make full use of human body parts characteristics such as human face characteristic, human body characteristic and head and shoulder characteristic, can extract more robust characteristic when not increasing the buffer memory, effectively promote the heavy rate of accuracy that removes, and be sheltered from to the face, human body (being the human body back to camera equipment) and the side direction human body (being the human body side to camera equipment) have better robustness, and personnel's heavy recall rate and validity are more excellent, and application scope is wider.
Referring to fig. 2, fig. 2 is a schematic flowchart illustrating another embodiment of a target re-identification method provided in the present application, the method including:
s201: and carrying out target detection processing on the image to be detected to obtain a detection result.
The video to be detected comprises an image to be detected, and the image to be detected is detected by adopting a target detection scheme in the related technology to obtain a detection result. Specifically, the detection result includes face position information and body position information of the snapshot target, the face position information is the position of the face, and the face position information includes coordinates of the upper left corner of a detection frame (denoted as a face detection frame) in which the face is located and coordinates of the lower right corner of the face detection frame; the human body position information is the position of the human body, and the human body position information comprises the coordinates of the upper left corner of a detection frame (marked as a human body detection frame) where the human body is located and the coordinates of the lower right corner of the human body detection frame.
S202: and tracking the snapshot target based on the detection result to obtain a tracking result.
After the detection result of the snap-shot target is obtained, a target tracking scheme in the related technology is adopted to track the detection result, and the tracking result of each snap-shot target is obtained, wherein the tracking result comprises related information of the snap-shot target, and the related information comprises an image to be detected where the snap-shot target is located, face position information and human body position information of the snap-shot target.
S203: and acquiring an image corresponding to the face position information of the snapshot target in the tracking result to obtain a first image to be evaluated, and acquiring an image corresponding to the human body position information of the snapshot target in the tracking result to obtain a second image to be evaluated.
After the tracking result of the snapshot target is obtained, an image (marked as a first image to be evaluated) corresponding to the face position information and an image (marked as a second image to be evaluated) corresponding to the human body position information are scratched out from all images to be detected in the tracking result, so that a snapshot image set is generated by using the first image to be evaluated and the second image to be evaluated. For example, assuming that the size of the image to be detected is 64 × 64, and the face position information is { (5,15), (10,20) }, the images of the 5 th to 10 th rows and 15 th to 20 th columns in the image to be detected are taken as the first image to be evaluated.
S204: and generating a snapshot image set based on the first image to be evaluated and the second image to be evaluated.
After the first image to be evaluated or the second image to be evaluated is acquired, the following scheme can be adopted to generate a snap shot image set:
A1) performing quality evaluation processing on the first image to be evaluated to obtain a face quality score; and screening out a first image from all the first images to be evaluated based on the face quality score.
And selecting a first image to be evaluated with the highest face quality score from all the first images to be evaluated to obtain a first image. Specifically, the face quality score may be determined according to the face sharpness and the face occlusion condition, for example: setting an occlusion score value for the face occlusion condition, and carrying out weighted summation on the face definition and the occlusion score value to obtain a face quality score; it is to be understood that the scheme in the related art may also be adopted to determine the face quality score, which is not limited herein.
A2) Performing quality evaluation processing on the second image to be evaluated to obtain a human body quality score; and screening out the human body image from all the second images to be evaluated based on the human body quality score.
The second image comprises a human body image and a head and shoulder image, and a human body quality evaluation scheme in the related technology can be adopted to score the human body in the second image to be evaluated to obtain a human body quality score. Performing gesture recognition processing on the second image to be evaluated by adopting a gesture recognition scheme in the correlation technique so as to obtain the gesture of the snapshot target in the second image to be evaluated; and respectively selecting a second image to be evaluated with the highest human quality score from all second images to be evaluated corresponding to each posture to obtain the human body image with the corresponding posture.
For example, assuming that the second images to be evaluated corresponding to the human body in the forward posture are denoted as I11 to I14, and the second images to be evaluated corresponding to the human body in the backward posture are denoted as I21 to I26, the second image to be evaluated with the highest score of the human body mass in the second images to be evaluated I11 to I14 is taken as the second image corresponding to the forward posture, and the second image to be evaluated with the highest score of the human body mass in the second images to be evaluated I21 to I26 is taken as the second image corresponding to the backward posture.
A3) Based on the human body image, a head-shoulder image is generated.
And (4) capturing the image of the head and the shoulder of the snapshot target from the human body image to obtain a head and shoulder image.
In an embodiment, an Identity Document (ID) may be set for a snapshot target, and for the same snapshot target, queue caching may be performed on all the human body position information, the human face position information, and the corresponding ID of the snapshot target, cache a first image to be evaluated (i.e., a first image) with the highest score of the human face quality, and cache an image (i.e., a second image) with the highest score of the human body quality in three postures (i.e., a forward posture, a lateral posture, and a backward posture). Meanwhile, a historical frame track of each snapshot target is maintained, the historical frame track comprises human body position information of the snapshot target in a historical frame image, and the historical frame image is an image before the current image to be detected in the video to be detected.
S205: and carrying out attribute identification processing on the snap-shot target in the second image to obtain attribute characteristics.
And processing the snap-shot target in the second image by adopting an attribute identification method in the related technology to obtain attribute characteristics. Specifically, if a second image exists in the snapshot image set, performing attribute identification processing on the second image; if at least two second images exist in the snapshot image set, performing attribute identification processing on each second image to obtain corresponding attribute characteristics, and then voting on the attribute characteristics by adopting a voting mechanism to obtain final attribute characteristics; or if at least two second images exist in the snapshot image set, calculating the highest value of the human body quality score of all the second images in the snapshot image set, and taking the attribute feature corresponding to the highest value as the final attribute feature.
Furthermore, information such as the color of the upper garment, the color of the lower garment, the age, the sex, the texture of the clothes and the like of the snap-shot target can be structurally described, and structural attribute labels and corresponding confidence coefficients of the snap-shot target are output. For example, the structured attribute labels are { white jacket, black under coat, 20, woman }, and the corresponding confidences are 0.86, 0.75, 0.92, and 0.95, respectively.
S206: responding to the fact that the first image and the second images corresponding to the multiple postures of the snapshot target exist in the snapshot image set, processing the face contained in the first image to obtain the face characteristics of the snapshot target, and processing the human body contained in the second images corresponding to the multiple postures to obtain the head and shoulder characteristics of the multiple postures and the human body characteristics of the multiple postures.
The first image is a first image to be evaluated with the highest face quality score in all first images to be evaluated, and the face features can be obtained by extracting the feature vector of the first image to be evaluated with the highest face quality score. It is understood that S206 is the same as S12 in the above embodiment, and is not described herein again.
S207: and fusing the head-shoulder characteristics of the plurality of postures to obtain multi-posture head-shoulder characteristics, and fusing the human body characteristics of the plurality of postures to obtain multi-posture human body characteristics.
S207 is the same as S13 in the above embodiment, and is not described again.
The method and the device extract and fuse the human body characteristics of the snapshot target in various postures such as the forward posture, the lateral posture, the backward posture and the like in real time, can obtain more robust multi-posture human body characteristics and dynamic information, overcome the problem of single-posture human body information unicity, and enrich human body characteristic information.
S208: and generating a re-recognition result based on the human face features, the multi-pose human body features, the multi-pose head and shoulder features and the attribute features.
The re-recognition result includes whether the snapshot target is a history snapshot target, and the following scheme can be adopted to generate the re-recognition result:
1) and acquiring the face quality score and the face integrity of the snapshot target.
The human face integrity can be obtained by a method in the related technology; such as: the integrity of the face can be determined by judging whether the face is shielded, and if the face is shielded, the integrity of the face is 0; if the face is not occluded, the face integrity is 1. It will be appreciated that the face integrity may also be determined by other schemes, such as: and calculating the ratio or the area of the human face which is shielded.
2) And judging whether the face quality score is greater than a first preset score threshold value or not, and judging whether the face integrity is greater than a first preset integrity threshold value or not.
The first preset scoring threshold and the first preset integrity threshold may be set according to experience or application requirements, and are not limited herein.
3) In response to the fact that the face quality score is larger than a first preset score threshold and the face integrity is larger than a first preset integrity threshold, calculating the similarity between the face features of the snapshot target and the face features of the historical snapshot target to obtain face similarity; if the face similarity is larger than the first preset similarity, determining that the re-recognition result is that the snapshot target is a historical snapshot target; and if the face similarity is smaller than or equal to the first preset similarity, determining that the re-recognition result is that the snapshot target is not the historical snapshot target.
4) And responding to the fact that the face quality score is smaller than or equal to a first preset score threshold value or the face integrity is smaller than or equal to a first preset integrity threshold value, and obtaining the human body quality score and the human body integrity of the snapshot target.
5) And judging whether the human body quality score is greater than a second preset score threshold value or not, and judging whether the human body integrity is greater than a second preset integrity threshold value or not.
6) And generating a re-recognition result based on the multi-pose human body characteristics and the attribute characteristics in response to the fact that the human body quality score is larger than a second preset score threshold and the human body integrity is larger than a second preset integrity threshold.
If the human body quality score is larger than a second preset score threshold and the human body integrity is larger than a second preset integrity threshold, generating a re-recognition result by adopting the following scheme:
calculating the similarity between the multi-pose human body characteristics of the snapshot target and the multi-pose human body characteristics of the historical snapshot target to obtain human body similarity; and calculating the similarity between the attribute characteristics of the snapshot target and the attribute characteristics of the historical snapshot target to obtain the attribute similarity.
Judging whether the human body similarity is greater than a second preset similarity or not, and judging whether the attribute similarity is greater than a third preset similarity or not; if the human body similarity is greater than the second preset similarity and the attribute similarity is greater than the third preset similarity, determining that the re-recognition result is that the snapshot target is a historical snapshot target; if the human body similarity is smaller than or equal to the second preset similarity, determining that the re-recognition result is that the snapshot target is not the historical snapshot target; and if the attribute similarity is less than or equal to the third preset similarity, determining that the re-recognition result is that the snapshot target is not the historical snapshot target.
7) Responding to the fact that the human body quality score is smaller than or equal to a second preset score threshold value or the human body integrity is smaller than or equal to a second preset integrity threshold value, calculating the similarity between the multi-pose head-shoulder features of the snapshot target and the multi-pose head-shoulder features of the historical snapshot target in the target database, and obtaining the head-shoulder similarity; judging whether the head-shoulder similarity is greater than a fourth preset similarity or not; if the head-shoulder similarity is greater than the fourth preset similarity, determining that the re-recognition result is that the snapshot target is a historical snapshot target; and if the head-shoulder similarity is less than or equal to the fourth preset similarity, determining that the re-recognition result is that the snapshot target is not the historical snapshot target.
According to the embodiment, the effectiveness of various features is determined according to the quality dimensions such as the face quality score, the face integrity, the body quality score and the body integrity, the low-quality face features and the low-quality body features are filtered, the feature comparison effect is improved, the error matching is reduced, and the duplicate removal accuracy rate is improved. Moreover, the head and shoulder characteristics of various postures are fused and compared, the effectiveness problem of the characteristics of a sheltered or incomplete human body can be solved, and the applicability is stronger.
S209: and responding to the re-recognition result that the snapshot target is not a historical snapshot target, and storing the human face feature, the multi-pose human body feature and the multi-pose head-shoulder feature of the snapshot target into a target database.
The relevant information of each snap-shot target can further comprise face features, face quality scores, body quality scores, multi-pose body features, multi-pose head and shoulder features and attribute features. According to the duplication elimination strategy, if the input snapshot target is different from all historical snapshot targets, the snapshot target is judged to be a non-repetitive person, and the related information of the snapshot target is added into a target database.
S210: and updating the target database based on the face features, multi-pose head-shoulder features and multi-pose human body features of the snapshot target in response to the re-recognition result that the snapshot target is a historical snapshot target.
If the snapshot target is the same as a certain historical snapshot target, the snapshot target is indicated to be a repeated target, and the target database can be updated by adopting the following scheme:
B1) and judging whether the face quality score of the snapshot target is larger than the face quality score of the historical snapshot target.
B2) And if the face quality score of the snapshot target is larger than the face quality score of the historical snapshot target, updating the face quality score of the historical snapshot target to the face quality score of the snapshot target, and updating the face characteristic of the historical snapshot target to the face characteristic of the snapshot target. And if the face quality score of the snapshot target is less than or equal to the face quality score of the historical snapshot target, not updating the corresponding face feature and the face quality score in the target database.
B3) Judging whether the human body quality score of the snapshot target is larger than the human body quality score of the historical snapshot target or not; if the human quality score of the snapshot target is larger than the human quality score of the historical snapshot target, updating the human quality score of the historical snapshot target to the human quality score of the snapshot target, updating the multi-posture human characteristics of the historical snapshot target to the multi-posture human characteristics of the snapshot target, updating the multi-posture head-shoulder characteristics of the historical snapshot target to the multi-posture head-shoulder characteristics of the snapshot target, and updating the attribute characteristics of the historical snapshot target to the attribute characteristics of the snapshot target. And if the human body quality score of the snapshot target is less than or equal to the human body quality score of the historical snapshot target, not updating the corresponding multi-pose human body characteristic, the human body quality score and the multi-pose head-shoulder characteristic in the target database.
According to the scheme, the target database is maintained according to the quality scores (including the face quality score and the human body quality score) of the snapshot targets, for the targets which are judged to be repeated, the face features and the human body features with better quality can be guaranteed to be updated into the target database by comparing the quality scores of the snapshot targets with the corresponding scores in the target database, so that the quality of the features in the target database is continuously improved, the comparison effect of the follow-up targets is guaranteed, and the system robustness is continuously improved along with the continuous lapse of time.
The embodiment makes full use of human body part characteristics such as human face and human body incidence relation, human face characteristics, human body characteristics, head and shoulder characteristics, and dress the human body and judge the dimension as the similarity with other structural attribute characteristics, multiple characteristics can complement each other, realize real-time personnel remove heavily to judge and distinguish, improved the application range of scheme by a wide margin, shelter from the human body to the face, human body, side direction and incomplete human body all have better robustness, personnel's remove heavily to recall rate and validity are more excellent.
Referring to fig. 3, fig. 3 is a schematic structural diagram of an embodiment of the object re-identification apparatus provided in the present application, the object re-identification apparatus 30 includes a memory 31 and a processor 32 connected to each other, the memory 31 is used for storing a computer program, and the computer program is used for implementing the object re-identification method in the foregoing embodiment when being executed by the processor 32.
Referring to fig. 4, fig. 4 is a schematic structural diagram of another embodiment of the object re-recognition apparatus provided in the present application, and the object re-recognition apparatus 40 includes an object capturing module 41, a feature extraction module 42, and a duplicate removal module 43.
The target snapshot module 41 is configured to process a video to be detected to obtain a snapshot image set for a snapshot target; specifically, the image collected by the front-end device (not shown in the figure) may be sent to the target snapshot module 41 at a frame rate of 8fps to 25fps, so that the target snapshot module 41 extracts and snapshots the human body and the human face appearing in the image.
In a specific embodiment, as shown in fig. 5, the target capture module 41 includes a target detection module 411, a target tracking module 412, and a processing module 413.
The target detection module 411 is configured to detect each frame of image to be detected in the video to be detected, so as to obtain a detection result, where the detection result includes a face of a snapshot target, a human body, or a head and a shoulder of the human body.
The target tracking module 412 is connected to the target detection module 411, and configured to perform tracking processing on the snapshot target based on the detection result to obtain a tracking result; specifically, the front frame image and the rear frame image of each snapshot target are subjected to space-time association, the face, the human body, the head and the shoulder of the snapshot target are subjected to association tracking, and a unique ID is generated.
The processing module 413 is connected to the target tracking module 412, and is configured to obtain an image corresponding to the face position information of the snapshot target in the tracking result, so as to obtain a first image to be evaluated; acquiring an image corresponding to the human body position information of the snap-shot target in the tracking result to obtain a second image to be evaluated; and generating a snapshot image set based on the first image to be evaluated and the second image to be evaluated.
Further, the processing module 413 includes a face quality scoring module 4131, a body quality scoring module 4132, and a snapshot module 4133.
The face quality scoring module 4131 is connected to the target tracking module 412, and is configured to perform quality evaluation processing on the first image to be evaluated to obtain a face quality score. Specifically, the face quality scoring module 4131 processes the image where the face is located in the data output by the target tracking module 412, and outputs the face quality score and information on whether the face is occluded.
The human quality scoring module 4132 is connected to the target tracking module 412, and is configured to perform quality evaluation processing on the second image to be evaluated to obtain a human quality score. Specifically, the human quality scoring module 4132 processes the image of the human body in the data output by the target tracking module 412, and outputs the human quality score, whether the human body is complete, and the orientation information of the human body, which is used to indicate whether the human body is in a forward posture, a lateral posture, or a back posture.
The snapshot module 4133 is connected to the face quality scoring module 4131 and the human quality scoring module 4132, and is configured to generate a first image and a second image based on the outputs of the human quality scoring module 4132 and the face quality scoring module 4131. Specifically, the first image may be a part of one of the second images corresponding to the plurality of gestures, or the first image does not belong to a part of any of the second images; the snapshot module 4133 is configured to screen out a first image from all first images to be evaluated based on the face quality score; screening out human body images from all second images to be evaluated based on the human body quality scores; based on the human body image, a head-shoulder image is generated.
Further, the snapshot module 4133 is configured to select a first image to be evaluated with the highest face quality score from all the first images to be evaluated, so as to obtain a first image. Acquiring the posture of a snapshot target in a second image to be evaluated; respectively selecting a second image to be evaluated with the highest human quality score from all second images to be evaluated corresponding to each posture to obtain a human image of the corresponding posture; and (4) capturing the image of the head and the shoulder of the snapshot target from the human body image to obtain a head and shoulder image.
By arranging the snapshot module 4133, when the snapshot target disappears or the trajectory of the snapshot target triggers line mixing and the like, the image with the highest quality score in the life cycle of the snapshot target can be obtained, and the complete human body image corresponding to the highest human body quality score and the corresponding head and shoulder images are obtained when the human body is in various postures such as a forward posture, a lateral posture and a back posture, so as to be used in subsequent feature extraction; the life cycle is a time period from the first time that the snapshot target appears in the target scene to the time that the snapshot target disappears from the target scene.
The feature extraction module 42 is connected to the target snapshot module 41, and is configured to, when a first image and second images corresponding to multiple poses of the snapshot target exist in the snapshot image set, process a human face included in the first image to obtain a human face feature of the snapshot target, and process a human body included in the second images corresponding to the multiple poses to obtain head-shoulder features of the multiple poses and human body features of the multiple poses; and fusing the head-shoulder characteristics of the plurality of postures to obtain multi-posture head-shoulder characteristics, and fusing the human body characteristics of the plurality of postures to obtain multi-posture human body characteristics. The feature extraction module 42 is further configured to, when the first image does not exist in the snapshot image set but the second images corresponding to the multiple postures of the snapshot target exist, process the human body included in the second images corresponding to the multiple postures to obtain the head and shoulder features of the multiple postures and the human body features of the multiple postures.
Further, the feature extraction module 42 is further configured to perform attribute identification processing on the snap-shot target in the second image to obtain an attribute feature; the attribute feature is used for representing attribute information of the snapshot target, and the attribute information comprises clothes color, age or gender. The feature extraction module 42 is configured to perform feature extraction processing on the head-shoulder image to obtain head-shoulder features; averaging the head and shoulder characteristics of all postures to obtain multi-posture head and shoulder characteristics; carrying out feature extraction processing on the human body image to obtain human body features; and averaging the human body characteristics of all postures to obtain the multi-posture human body characteristics.
By arranging the feature extraction module 42, the relevant features of the snap-shot target in the life cycle of the snap-shot target with the same ID can be extracted, and the relevant features include human face features, human body part features (such as head and shoulder features, foot features, leg features or hand features) or attribute features, which are used as the comparison basis of the duplication elimination module 43.
The duplicate removal module 43 is connected to the feature extraction module 42, and configured to generate a duplicate recognition result based on the human face feature, the multi-pose human body feature, and the multi-pose head-shoulder feature when the first image and the second images corresponding to the multiple poses of the snapshot target exist in the snapshot image set, where the duplicate recognition result includes whether the snapshot target is a historical snapshot target; when the first image does not exist in the snapshot image set but the second images corresponding to the multiple postures of the snapshot target exist, a re-recognition result is generated based on the multi-posture human body characteristic and the multi-posture head-shoulder characteristic.
In a specific embodiment, the duplication elimination module 43 is further configured to generate a duplication recognition result based on the human face features, the multi-pose human body features, the multi-pose head-shoulder features, and the attribute features; or generating a re-recognition result based on the multi-pose human body features, the multi-pose head-shoulder features and the attribute features. Specifically, when the input of the duplication elimination module 43 is the face features, multi-pose human body features, multi-pose head and shoulder features, and attribute features of the snapshot target, the duplication elimination module 43 compares the features with the similarity of the target database maintained in real time, and if the similarity meets the requirement, the same snapshot target is considered, otherwise, the new snapshot target is considered.
The duplication elimination module 43 is used for acquiring the face quality score and the face integrity of the snapshot target; if the face quality score is larger than a first preset score threshold and the face integrity is larger than a first preset integrity threshold, calculating the similarity between the face features of the snapshot target and the face features of the historical snapshot target to obtain the face similarity; judging whether the face similarity is greater than a first preset similarity or not; if the face similarity is larger than the first preset similarity, determining that the re-recognition result is that the snapshot target is a historical snapshot target; if not, determining that the re-recognition result is that the snapshot target is not the historical snapshot target. If the human body quality score is smaller than or equal to a second preset score threshold value, or the human body integrity is smaller than or equal to a second preset integrity threshold value, calculating the similarity between the multi-pose head-shoulder features of the snapshot target and the multi-pose head-shoulder features of the historical snapshot target to obtain head-shoulder similarity; judging whether the head-shoulder similarity is greater than a fourth preset similarity or not; if the head-shoulder similarity is greater than the fourth preset similarity, determining that the re-recognition result is that the snapshot target is a historical snapshot target; and if the head-shoulder similarity is less than or equal to the fourth preset similarity, determining that the re-recognition result is that the snapshot target is not the historical snapshot target. If the face quality score is smaller than or equal to a first preset score threshold value, or the face integrity is smaller than or equal to a first preset integrity threshold value, acquiring the human body quality score and the human body integrity of the snapshot target; and if the human body quality score is larger than a second preset score threshold value and the human body integrity is larger than a second preset integrity threshold value, generating a re-recognition result based on the multi-pose human body characteristics and the attribute characteristics.
Further, the duplicate removal module 43 is configured to calculate similarity between the multi-pose human body features of the snapshot target and the multi-pose human body features of the historical snapshot target, so as to obtain human body similarity; calculating the similarity between the attribute characteristics of the snapshot target and the attribute characteristics of the historical snapshot target to obtain the attribute similarity; if the human body similarity is greater than the second preset similarity and the attribute similarity is greater than the third preset similarity, determining that the re-recognition result is that the snapshot target is a historical snapshot target; and if the human body similarity is less than or equal to the second preset similarity or the attribute similarity is less than or equal to the third preset similarity, determining that the re-identification result is that the snapshot target is not the historical snapshot target.
With continued reference to fig. 4, the target re-recognition apparatus 40 further includes a base update management module 44, the base update management module 44 is connected to the duplication elimination module 43, and the base update management module 44 is configured to determine whether to update the target database based on the face quality score, the body quality score, the face feature, the multi-pose body feature, the multi-pose head and shoulder feature, and the attribute feature. Specifically, if the re-recognition result is that the snapshot target is not a historical snapshot target, storing the human face features, multi-pose human body features and multi-pose head-shoulder features of the snapshot target into a target database; and if the re-recognition result is that the snapshot target is a historical snapshot target, updating the target database based on the human face features, multi-pose head-shoulder features and multi-pose human body features of the snapshot target.
Further, the base library update management module 44 is configured to determine whether the face quality score of the snapshot target is greater than the face quality score of the historical snapshot target; and if the face quality score of the snapshot target is larger than the face quality score of the historical snapshot target, updating the face quality score of the historical snapshot target to the face quality score of the snapshot target, and updating the face characteristic of the historical snapshot target to the face characteristic of the snapshot target. Judging whether the human body quality score of the snapshot target is larger than the human body quality score of the historical snapshot target or not; if the human quality score of the snapshot target is larger than the human quality score of the historical snapshot target, the human quality score of the historical snapshot target is updated to be the human quality score of the snapshot target, the multi-pose human characteristics of the historical snapshot target are updated to be the multi-pose human characteristics of the snapshot target, the multi-pose head-shoulder characteristics of the historical snapshot target are updated to be the multi-pose head-shoulder characteristics of the snapshot target, and the attribute characteristics of the historical snapshot target are updated to be the attribute characteristics of the snapshot target.
The embodiment combines human face features, multi-pose human body features, multi-pose head and shoulder features and attribute features to perform multi-dimensional personnel identification, mutual complementation is achieved, the deduplication recall rate is improved, good robustness is achieved for human bodies, back bodies, lateral human bodies and incomplete human bodies, the problem that the algorithm fails due to the fact that the human faces are sheltered or depend on the features singly when no human face is found can be solved, the deduplication recall rate and effectiveness of personnel are better, and the use range of the scheme is greatly enlarged.
Referring to fig. 6, fig. 6 is a schematic structural diagram of an embodiment of a computer-readable storage medium 60 provided in the present application, where the computer-readable storage medium 60 is used for storing a computer program 61, and when the computer program 61 is executed by a processor, the computer program is used for implementing the object re-identification method in the foregoing embodiment.
The computer readable storage medium 60 may be a server, a usb disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, and various media capable of storing program codes.
In the several embodiments provided in the present application, it should be understood that the disclosed method and apparatus may be implemented in other manners. For example, the above-described apparatus embodiments are merely illustrative, and for example, a division of modules or units is merely a logical division, and an actual implementation may have another division, for example, a plurality of units or components may be combined or integrated into another system, or some features may be omitted, or not executed.
The units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one position, or may be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the embodiment.
In addition, functional units in the embodiments of the present application may be integrated into one processing unit, or each unit may exist alone physically, or two or more units may be integrated into one unit. The integrated unit can be realized in a form of hardware, and can also be realized in a form of a software functional unit.
If the technical scheme of the application relates to personal information, a product applying the technical scheme of the application clearly informs personal information processing rules before processing the personal information, and obtains personal independent consent. If the technical scheme of the application relates to sensitive personal information, a product applying the technical scheme of the application obtains individual consent before processing the sensitive personal information, and simultaneously meets the requirement of 'express consent'. For example, at a personal information collection device such as a camera, a clear and significant identifier is set to inform that the personal information collection range is entered, the personal information is collected, and if the person voluntarily enters the collection range, the person is considered as agreeing to collect the personal information; or on the device for processing the personal information, under the condition of informing the personal information processing rule by using obvious identification/information, obtaining personal authorization by modes of popping window information or asking a person to upload personal information of the person by himself, and the like; the personal information processing rule may include information such as a personal information processor, a personal information processing purpose, a processing method, and a type of personal information to be processed.
The above embodiments are merely examples, and not intended to limit the scope of the present application, and all modifications, equivalents, and flow charts using the contents of the specification and drawings of the present application, or those directly or indirectly applied to other related arts, are included in the scope of the present application.

Claims (19)

1. A target re-identification method is characterized by comprising the following steps:
processing a video to be detected to obtain a snapshot image set aiming at a snapshot target;
responding to a first image and second images corresponding to a plurality of postures of the snapshot target in the snapshot image set, processing a human face contained in the first image to obtain human face characteristics of the snapshot target, and processing a human body contained in the second images corresponding to the plurality of postures to obtain head and shoulder characteristics of the plurality of postures and human body characteristics of the plurality of postures;
fusing the head-shoulder characteristics of the plurality of postures to obtain multi-posture head-shoulder characteristics, and fusing the human body characteristics of the plurality of postures to obtain multi-posture human body characteristics;
generating a re-recognition result based on the human face features, the multi-pose human body features and the multi-pose head and shoulder features, wherein the re-recognition result comprises whether the snapshot target is a historical snapshot target.
2. The object re-identification method according to claim 1, further comprising:
responding to second images corresponding to a plurality of postures of the snapshot target but without the first image in the snapshot image set, and processing human bodies contained in the second images corresponding to the plurality of postures to obtain head and shoulder features of the plurality of postures and human body features of the plurality of postures;
and generating the re-recognition result based on the multi-pose human body features and the multi-pose head-shoulder features.
3. The object re-identification method according to claim 2, wherein after the step of processing the video to be detected to obtain the snap-shot image set for the snap-shot object, the method comprises:
performing attribute identification processing on the snap-shot target in the second image to obtain attribute characteristics;
the step of generating a re-recognition result based on the face features, the multi-pose body features, and the multi-pose head-shoulder features includes:
generating the re-recognition result based on the face features, the multi-pose human body features, the multi-pose head-shoulder features and the attribute features;
the step of generating the re-recognition result based on the multi-pose human body features and the multi-pose head-shoulder features comprises:
and generating the re-recognition result based on the multi-pose human body features, the multi-pose head-shoulder features and the attribute features.
4. The method for re-identifying the target as claimed in claim 3, wherein the video to be detected comprises an image to be detected, and the step of processing the video to be detected to obtain a snapshot image set for the snapshot target comprises:
carrying out target detection processing on the image to be detected to obtain a detection result;
tracking the snapshot target based on the detection result to obtain a tracking result, wherein the tracking result comprises related information of the snapshot target, the related information comprises face position information and human body position information of the snapshot target, the face position information is the position of the face, and the human body position information is the position of the human body;
acquiring an image corresponding to the face position information of the snap-shot target in the tracking result to obtain a first image to be evaluated;
acquiring an image corresponding to the human body position information of the snapshot target in the tracking result to obtain a second image to be evaluated;
and generating the snapshot image set based on the first image to be evaluated and the second image to be evaluated.
5. The object re-recognition method according to claim 4, wherein the second image includes a human body image and a head-shoulder image, and the step of generating the snap-shot image set based on the first image to be evaluated and the second image to be evaluated includes:
performing quality evaluation processing on the first image to be evaluated to obtain a face quality score;
performing quality evaluation processing on the second image to be evaluated to obtain a human body quality score;
screening the first image from all the first images to be evaluated based on the face quality score;
screening the human body image from all the second images to be evaluated based on the human body quality score;
generating the head and shoulder image based on the human body image;
the step of processing the human body included in the second image corresponding to the plurality of postures to obtain the head and shoulder features of the plurality of postures and the human body features of the plurality of postures comprises:
carrying out feature extraction processing on the human body image to obtain the human body features;
and performing feature extraction processing on the head and shoulder image to obtain the head and shoulder features.
6. The object re-recognition method of claim 5, wherein the step of screening out the first image from all the first images to be evaluated based on the face quality score comprises:
and selecting a first image to be evaluated with the highest face quality score from all the first images to be evaluated to obtain the first image.
7. The object re-recognition method according to claim 5, wherein the step of screening out the human body image from all the second images to be evaluated based on the human body quality score comprises:
acquiring the posture of a snapshot target in the second image to be evaluated;
and respectively selecting a second image to be evaluated with the highest human quality score from all the second images to be evaluated corresponding to each posture to obtain the human body image with the corresponding posture.
8. The object re-recognition method according to claim 5, wherein the step of generating the head-shoulder image based on the human body image comprises:
and intercepting the image of the head and the shoulder of the snapshot target from the human body image to obtain the head and shoulder image.
9. The object re-recognition method of claim 3, wherein the step of generating the re-recognition result based on the face features, the multi-pose human body features, the multi-pose head-shoulder features, and the attribute features comprises:
acquiring the face quality score and the face integrity of the snapshot target;
responding to the fact that the face quality score is larger than a first preset score threshold value and the face integrity is larger than a first preset integrity threshold value, calculating the similarity between the face features of the snapshot target and the face features of the historical snapshot target, and obtaining the face similarity;
judging whether the face similarity is larger than a first preset similarity or not;
if yes, determining that the re-recognition result is that the snapshot target is the historical snapshot target;
if not, determining that the re-recognition result is that the snapshot target is not the historical snapshot target.
10. The method of claim 9, wherein the step of generating the re-recognition result based on the facial features, the multi-pose body features, the multi-pose head-shoulder features, and the attribute features further comprises:
responding to the fact that the face quality score is smaller than or equal to the first preset score threshold value or the face integrity is smaller than or equal to the first preset integrity threshold value, and obtaining the human body quality score and the human body integrity of the snapshot target;
and generating the re-recognition result based on the multi-posture human body characteristics and the attribute characteristics in response to the fact that the human body quality score is larger than a second preset score threshold and the human body integrity is larger than a second preset integrity threshold.
11. The object re-recognition method according to claim 10, wherein the step of generating the re-recognition result based on the multi-pose human features and the attribute features comprises:
calculating the similarity between the multi-pose human body characteristics of the snapshot target and the multi-pose human body characteristics of the historical snapshot target to obtain human body similarity;
calculating the similarity between the attribute characteristics of the snapshot target and the attribute characteristics of the historical snapshot target to obtain the attribute similarity;
responding to the fact that the human body similarity is larger than a second preset similarity and the attribute similarity is larger than a third preset similarity, and determining that the re-recognition result is that the snapshot target is the historical snapshot target;
and determining that the re-recognition result is that the snapshot target is not the historical snapshot target in response to the fact that the human body similarity is smaller than or equal to a second preset similarity or the attribute similarity is smaller than or equal to a third preset similarity.
12. The object re-recognition method of claim 10, wherein the step of generating the re-recognition result based on the multi-pose human features and the attribute features further comprises:
responding to the fact that the human body quality score is smaller than or equal to the second preset score threshold value or the human body integrity is smaller than or equal to the second preset integrity threshold value, calculating the similarity between the multi-pose head-shoulder characteristics of the snapshot target and the multi-pose head-shoulder characteristics of the historical snapshot target, and obtaining the head-shoulder similarity;
judging whether the head-shoulder similarity is greater than a fourth preset similarity or not;
if so, determining that the re-recognition result is that the snapshot target is the historical snapshot target;
if not, determining that the re-recognition result is that the snapshot target is not the historical snapshot target.
13. The object re-recognition method of claim 3,
the first image is a part of one of the second images corresponding to the plurality of postures, the attribute feature is used for representing attribute information of the snapshot target, and the attribute information comprises clothes color, age or gender.
14. The object re-identification method according to claim 1, characterized in that the method further comprises:
responding to the re-recognition result that the snapshot target is not the historical snapshot target, and storing the human face feature, the multi-pose human body feature and the multi-pose head and shoulder feature of the snapshot target into a target database;
and in response to the re-recognition result that the snapshot target is the historical snapshot target, updating the target database based on the human face features, multi-pose head-shoulder features and multi-pose human body features of the snapshot target.
15. The method of claim 14, wherein the step of updating the target database based on the face features, multi-pose head-shoulder features, and multi-pose body features of the snap-shot target comprises:
judging whether the face quality score of the snapshot target is larger than the face quality score of the historical snapshot target or not;
if so, updating the face quality score of the historical snap-shot target to the face quality score of the snap-shot target, and updating the face feature of the historical snap-shot target to the face feature of the snap-shot target.
16. The method of claim 14, wherein the step of updating the target database based on the face features, multi-pose head-shoulder features, and multi-pose body features of the snap-shot target further comprises:
judging whether the human body quality score of the snapshot target is larger than the human body quality score of the historical snapshot target or not;
if so, updating the human quality score of the historical snap-shot target to the human quality score of the snap-shot target, updating the multi-posture human characteristics of the historical snap-shot target to the multi-posture human characteristics of the snap-shot target, updating the multi-posture head-shoulder characteristics of the historical snap-shot target to the multi-posture head-shoulder characteristics of the snap-shot target, and updating the attribute characteristics of the historical snap-shot target to the attribute characteristics of the snap-shot target.
17. The object re-recognition method according to claim 1, wherein the step of fusing the head-shoulder features of the plurality of postures to obtain multi-posture head-shoulder features comprises:
averaging the head-shoulder characteristics of all the postures to obtain the multi-posture head-shoulder characteristics;
the step of fusing the human body characteristics of the plurality of postures to obtain the human body characteristics of the plurality of postures comprises the following steps:
and averaging the human body characteristics of all the postures to obtain the multi-posture human body characteristics.
18. An object re-recognition apparatus comprising a memory and a processor connected to each other, wherein the memory is used for storing a computer program, which when executed by the processor is used for implementing the object re-recognition method according to any one of claims 1-17.
19. A computer-readable storage medium for storing a computer program, wherein the computer program, when executed by a processor, is configured to implement the object re-identification method of any one of claims 1-17.
CN202210689509.6A 2022-06-17 2022-06-17 Object re-recognition method, object re-recognition apparatus, and computer-readable storage medium Active CN114783037B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210689509.6A CN114783037B (en) 2022-06-17 2022-06-17 Object re-recognition method, object re-recognition apparatus, and computer-readable storage medium

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210689509.6A CN114783037B (en) 2022-06-17 2022-06-17 Object re-recognition method, object re-recognition apparatus, and computer-readable storage medium

Publications (2)

Publication Number Publication Date
CN114783037A true CN114783037A (en) 2022-07-22
CN114783037B CN114783037B (en) 2022-11-22

Family

ID=82421855

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210689509.6A Active CN114783037B (en) 2022-06-17 2022-06-17 Object re-recognition method, object re-recognition apparatus, and computer-readable storage medium

Country Status (1)

Country Link
CN (1) CN114783037B (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115631509A (en) * 2022-10-24 2023-01-20 智慧眼科技股份有限公司 Pedestrian re-identification method and device, computer equipment and storage medium
CN117152826A (en) * 2023-11-01 2023-12-01 武汉中科通达高新技术股份有限公司 Real-time cross-mirror tracking method based on target tracking and anomaly detection

Citations (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2011151997A1 (en) * 2010-06-01 2011-12-08 Sony Corporation Information processing apparatus and method and program
CN105139040A (en) * 2015-10-13 2015-12-09 商汤集团有限公司 Queuing state information detection method and system thereof
CN109657533A (en) * 2018-10-27 2019-04-19 深圳市华尊科技股份有限公司 Pedestrian recognition methods and Related product again
CN110796079A (en) * 2019-10-29 2020-02-14 深圳龙岗智能视听研究院 Multi-camera visitor identification method and system based on face depth features and human body local depth features
CN110929711A (en) * 2019-11-15 2020-03-27 智慧视通(杭州)科技发展有限公司 Method for automatically associating identity information and shape information applied to fixed scene
WO2020249025A1 (en) * 2019-06-12 2020-12-17 杭州海康威视数字技术股份有限公司 Identity information determining method and apparatus, and storage medium
CN112488071A (en) * 2020-12-21 2021-03-12 重庆紫光华山智安科技有限公司 Method, device, electronic equipment and storage medium for extracting pedestrian features
CN113065491A (en) * 2021-04-12 2021-07-02 成都睿码科技有限责任公司 Multi-shooting-based passenger flow statistical method and system
CN113111795A (en) * 2021-04-18 2021-07-13 上海资汇信息科技有限公司 Positioning and tracking device and method based on pedestrian re-identification
CN113128454A (en) * 2021-04-30 2021-07-16 北京智通东方软件科技有限公司 Video processing method and device, storage medium and electronic equipment
CN114357216A (en) * 2021-12-10 2022-04-15 浙江大华技术股份有限公司 Portrait gathering method and device, electronic equipment and storage medium
CN114627526A (en) * 2022-02-14 2022-06-14 厦门瑞为信息技术有限公司 Fusion duplicate removal method and device based on multi-camera snapshot image and readable medium

Patent Citations (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2011151997A1 (en) * 2010-06-01 2011-12-08 Sony Corporation Information processing apparatus and method and program
CN105139040A (en) * 2015-10-13 2015-12-09 商汤集团有限公司 Queuing state information detection method and system thereof
CN109657533A (en) * 2018-10-27 2019-04-19 深圳市华尊科技股份有限公司 Pedestrian recognition methods and Related product again
WO2020249025A1 (en) * 2019-06-12 2020-12-17 杭州海康威视数字技术股份有限公司 Identity information determining method and apparatus, and storage medium
CN110796079A (en) * 2019-10-29 2020-02-14 深圳龙岗智能视听研究院 Multi-camera visitor identification method and system based on face depth features and human body local depth features
CN110929711A (en) * 2019-11-15 2020-03-27 智慧视通(杭州)科技发展有限公司 Method for automatically associating identity information and shape information applied to fixed scene
CN112488071A (en) * 2020-12-21 2021-03-12 重庆紫光华山智安科技有限公司 Method, device, electronic equipment and storage medium for extracting pedestrian features
CN113065491A (en) * 2021-04-12 2021-07-02 成都睿码科技有限责任公司 Multi-shooting-based passenger flow statistical method and system
CN113111795A (en) * 2021-04-18 2021-07-13 上海资汇信息科技有限公司 Positioning and tracking device and method based on pedestrian re-identification
CN113128454A (en) * 2021-04-30 2021-07-16 北京智通东方软件科技有限公司 Video processing method and device, storage medium and electronic equipment
CN114357216A (en) * 2021-12-10 2022-04-15 浙江大华技术股份有限公司 Portrait gathering method and device, electronic equipment and storage medium
CN114627526A (en) * 2022-02-14 2022-06-14 厦门瑞为信息技术有限公司 Fusion duplicate removal method and device based on multi-camera snapshot image and readable medium

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
JIA LI ET AL.: "Multi-pose learning based head-shoulder re-identification", 《2018 IEEE CONFERENCE ON MULTIMEDIA INFORMATION PROCESSING AND RETRIEVAL(MIPR)》 *
郑舟恒等: "基于测度矩阵正则化的行人重识别算法", 《计算机工程与设计》 *

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115631509A (en) * 2022-10-24 2023-01-20 智慧眼科技股份有限公司 Pedestrian re-identification method and device, computer equipment and storage medium
CN117152826A (en) * 2023-11-01 2023-12-01 武汉中科通达高新技术股份有限公司 Real-time cross-mirror tracking method based on target tracking and anomaly detection
CN117152826B (en) * 2023-11-01 2024-03-22 武汉中科通达高新技术股份有限公司 Real-time cross-mirror tracking method based on target tracking and anomaly detection

Also Published As

Publication number Publication date
CN114783037B (en) 2022-11-22

Similar Documents

Publication Publication Date Title
Vishnu et al. Human fall detection in surveillance videos using fall motion vector modeling
CN114783037B (en) Object re-recognition method, object re-recognition apparatus, and computer-readable storage medium
US7848548B1 (en) Method and system for robust demographic classification using pose independent model from sequence of face images
KR101764845B1 (en) A video surveillance apparatus for removing overlap and tracking multiple moving objects and method thereof
US11527000B2 (en) System and method for re-identifying target object based on location information of CCTV and movement information of object
CN109426785B (en) Human body target identity recognition method and device
Liu et al. Real-time human detection and tracking in complex environments using single RGBD camera
Merad et al. Tracking multiple persons under partial and global occlusions: Application to customers’ behavior analysis
CN111091025B (en) Image processing method, device and equipment
CN110781733B (en) Image duplicate removal method, storage medium, network equipment and intelligent monitoring system
US20170249670A1 (en) Targeted advertising and facial extraction and analysis
CN109033985B (en) Commodity identification processing method, device, equipment, system and storage medium
CN112016353A (en) Method and device for carrying out identity recognition on face image based on video
CN111652331A (en) Image recognition method and device and computer readable storage medium
CN108830204B (en) Method for detecting abnormality in target-oriented surveillance video
CN109146913B (en) Face tracking method and device
CN113920585A (en) Behavior recognition method and device, equipment and storage medium
Denman et al. Identifying customer behaviour and dwell time using soft biometrics
CN112001280A (en) Real-time online optimization face recognition system and method
CN112749605A (en) Identity recognition method, system and equipment
CN116071569A (en) Image selection method, computer equipment and storage device
CN111738059A (en) Non-sensory scene-oriented face recognition method
CN115482569A (en) Target passenger flow statistical method, electronic device and computer readable storage medium
CN113657169B (en) Gait recognition method, device and system and computer readable storage medium
WO2022134916A1 (en) Identity feature generation method and device, and storage medium

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant