Method and apparatus for performing person re-identification
By updating the person re-identification database in real time and selectively updating registered vectors using state information, the problem of decreased identification accuracy caused by state changes in the person re-identification system is solved, and the system's identification ability under drastic changes is improved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- REALTEK SEMICON CORP
- Filing Date
- 2021-11-29
- Publication Date
- 2026-06-05
Smart Images
Figure CN116206249B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to Person Re-Identification, and more particularly to a method and related apparatus for managing the registration and updating of a person re-identification database based on the person's status information, thereby performing person re-identification. Background Technology
[0002] Generally, applications using deep learning networks for identification, such as person re-identification, typically involve two processes: identity registration and database comparison. In the identity registration process, after obtaining a source image containing a person, the deep learning network outputs a feature vector representing that person and registers this vector, along with the person's identity (either user-inputted or automatically assigned by the system), in the database, establishing an association to complete the registration. In the database comparison process, when the person reappears in the source image, the deep learning network generates a new feature vector and compares it with all registered vectors in the database. If the distance between this vector and a registered vector in the database is less than a threshold, the person is deemed to possess the identity corresponding to that vector. Therefore, if the person's feature vector in the database cannot accurately reflect or represent the person's status, identification will fail. In other words, the control and management of identity registration directly affects the accuracy of identification. Summary of the Invention
[0003] This invention proposes a method and related apparatus for managing the updating and registration of a person re-identification database based on person state information, thereby enabling person re-identification. Specifically, to track changes in person state, this invention updates the person re-identification database in real time. Once a person feature vector calculated from the source image successfully matches a registered vector in the re-identification database, the current person feature vector is used to update the registered vector in the re-identification database. Furthermore, to ensure the re-identification database covers all possible changes in person state, this invention also references the current person state determined from the source image when registering and updating the database. The current person feature vector is only used to update registered vectors in the re-identification database that are similar to or related to the current person state. In this way, the person re-identification database can retain registered vectors that differ significantly from the current person state, ensuring accurate person identification even when the person state changes drastically.
[0004] An embodiment of the present invention provides a method for re-identifying a person, the method comprising: obtaining a person feature vector based on a captured image containing a person; obtaining state information based on the state of the person in the captured image; matching the person feature vector with a plurality of registered person feature vectors in a re-identification database; when the person feature vector successfully matches a first registered person feature vector among the plurality of registered person feature vectors, identifying the person as a first person identity corresponding to the first registered person feature vector; and selectively updating one of the first registered person feature vector and at least one second registered person feature vector corresponding to the first person identity using the person feature vector based on the state information.
[0005] An embodiment of the present invention provides an apparatus for performing person re-identification, the apparatus comprising: a storage unit and a processing unit. The storage unit stores program code. The processing unit executes the program code such that the processing unit performs the following operations: obtaining a person feature vector based on a captured image containing a person; obtaining state information based on the state of the person in the captured image; matching the person feature vector with a plurality of registered person feature vectors in a re-identification database; when the person feature vector successfully matches a first registered person feature vector among the plurality of registered person feature vectors, identifying the person as a first person identity corresponding to the first registered person feature vector; and selectively updating one of the first registered person feature vector and at least one second registered person feature vector corresponding to the first person identity using the person feature vector based on the state information. Attached Figure Description
[0006] Figure 1 An architecture diagram of a person re-identification system according to an embodiment of the present invention is shown.
[0007] Figure 2 This illustration depicts multiple registered person feature vectors stored in the re-identification database for a specific person in one embodiment of the present invention.
[0008] Figure 3 The diagram illustrates multiple registered person feature vectors stored in the re-identification database for a specific person in another embodiment of the present invention.
[0009] Figure 4 This illustration shows an example of the update process for the re-identification database in an embodiment of the present invention.
[0010] Figure 5 The diagram illustrates the architecture of a person re-identification method according to an embodiment of the present invention.
[0011] Figure 6The illustration depicts a possible implementation architecture of the person re-identification system in an embodiment of the present invention.
[0012] [Symbol Explanation]
[0013] 10 Image Capture System
[0014] 100 People Re-identification System
[0015] 110 Object Detection Module
[0016] 120 Re-identification Module
[0017] 121 Feature Vector Calculation Module
[0018] 122 State Calculation Module
[0019] 123 Database Comparison and Update Module
[0020] 124 Re-identification Database
[0021] Image source: IMG_S
[0022] Image capture using IMG_EXT, IMG_EXT_1, and IMG_EXT_2
[0023] VC_IN, VC_IN_1, VC_IN_2: Character Feature Vectors
[0024] R_Info status information
[0025] ID_Info Recognition Result
[0026] ID_1~ID_N Person Identity
[0027] VCreg_k_1~VCreg_k_m are the feature vectors of registered individuals.
[0028] S310~S350 Steps Detailed Implementation
[0029] Numerous specific details are described below to provide the reader with a thorough understanding of embodiments of the invention. However, those skilled in the art will appreciate how the invention can be implemented in the absence of one or more specific details, or by utilizing other methods, components, or materials. In other instances, well-known structures, materials, or operations are not shown or described in detail to avoid obscuring the core concepts of the invention.
[0030] The phrase "in one embodiment" in this specification means that a particular feature, structure, or characteristic described in that embodiment may be included in at least one embodiment of the invention. Therefore, the phrase "in one embodiment" appearing throughout this specification does not necessarily mean the same embodiment. Furthermore, the aforementioned particular features, structures, or characteristics may be combined in any suitable form in one or more embodiments.
[0031] An embodiment of the present invention provides a person re-identification system, please refer to... Figure 1 The architecture is shown. The human re-identification system 100 includes an object detection module 110 and a person re-identification module 120. Images generated by the image capturing system 10 are provided to the object detection module 110. In some embodiments, the image capturing system 10 may consist of one or more cameras. When the image capturing system 10 consists of multiple cameras, they may be located in different or adjacent locations, such as streets or indoor spaces. Furthermore, the image capturing system 10 can output a temporally continuous image sequence to the object detection module 110. When the object detection module 110 receives a source image IMG_S, it will perceive the human figure outline of person CK within it, thereby determining the presence of person CK, and extract an extracted image IMG_EXT containing person CK from the source image IMG_S. The extracted image IMG_EXT is provided to the re-identification module 120. The re-identification module 120 includes a feature vector calculation module 121, a state calculation module 122, a database comparison and update module 123, and a re-identification database 124, which are used to realize person re-identification.
[0032] The feature vector calculation module 121 is used to convert the human figure in the captured image IMG_EXT into a human feature vector VC_IN, where the human feature vector VC_IN may reflect the human figure CK's (but is not limited to) body shape, posture, clothing color, etc. The state calculation module 122 is used to identify the state of the human figure CK in the captured image IMG_EXT, thereby generating state information R_Info. In one embodiment of the present invention, the state information R_Info can be the upper body proportion, that is, the proportion of the upper body portion of the human figure in the captured image IMG_EXT relative to the whole. The state calculation module 122 can perform segmentation processing on the human figure in the captured image IMG_EXT to determine key parts such as the upper body and lower body, thereby calculating the upper body proportion. In other embodiments of the present invention, the state information R_Info can also be the rotation angle of the human figure's front view (i.e., the rotation angle of the human figure's front view relative to the center of the captured image IMG_EXT). Furthermore, the state information R_Info can also be the occlusion rate of the person in the image IMG_EXT (i.e., the ratio of the part of the person's body that is occluded by an object to the whole body), or it can be a parameter indicating the person's posture. In addition, in some embodiments of the present invention, the state information R_Info can also be a parameter composed of one or more of the aforementioned state information, so as to better reflect all the elements of the person's state.
[0033] The database comparison and update module 123 compares the person feature vector VC_IN with all registered person feature vectors in the re-identification database 124. The re-identification database 124 records person feature vectors corresponding to multiple different person identities ID_1 to ID_N. Furthermore, for a single person identity, the re-identification database 124 records multiple sets of related person feature vectors. For example, the re-identification database 124 may record m sets of registered person feature vectors VCreg_k_1 to VCreg_k_m for person identity ID_k. Please refer to [reference needed]. Figure 2This embodiment illustrates a scenario where m=3, and the database comparison and update module 123 uses the upper body proportion of the captured image IMG_EXT as the basis for updating and registration. Under this premise, the re-identification database 125 will store three different registered person feature vectors for the person's identity ID_k: vector VCreg_k_1 corresponding to the image from the person's head to near the waist, vector VCreg_k_2 corresponding to the image from the person's head to near the knees, with the lower body obscured, and vector VCreg_k_3 corresponding to the image from the person's head to above the knees. These three sets of registered person feature vectors VCreg_k_1 to VCreg_k_3 each correspond to different numerical ranges of upper body proportion. For example, the registered person feature vector VCreg_k_1 may correspond to an upper body proportion of 60-90%, the registered person feature vector VCreg_k_2 may correspond to an upper body proportion of 30-60%, and the registered person feature vector VCreg_k_3 may correspond to an upper body proportion of 0-30%. Going forward, the database comparison and update module 123 will use update thresholds TH1 (30%) and TH2 (60%) related to the upper body proportion to determine how to update the database comparison and update module 123 (details to be described later). Furthermore, Figure 3 The diagram illustrates the case where m=3, and the database comparison and update module 123 uses the rotation angle of the person's front view in the captured image IMG_EXT (i.e., the rotation angle of the person's front view relative to the center of the captured image IMG_EXT) as the basis for updating and registration. Under this premise, the re-identification database 125 will have three different registered person feature vectors stored for the person's identity ID_k: VCreg_k_1 corresponding to images with a front view rotation angle of 0-60 degrees, VCreg_k_2 corresponding to images with a front view rotation angle of 60-120 degrees, and VCreg_k_3 corresponding to images with a front view rotation angle of 120-180 degrees. Subsequently, the database comparison and update module 123 will use the relevant angle thresholds TH1 (60 degrees) and TH2 (120 degrees) of the front view rotation angle to determine how to update the database comparison and update module 123 (details to be described later).
[0034] It should be noted that in the above and subsequent descriptions, the inventive concept will continue to be explained with m=3, but this is not a limitation of the invention. In other embodiments of the invention, the re-identification database 124 may store more or fewer registered person feature vectors for a single person identity. These more or fewer registered person vectors may correspond to narrower or wider numerical ranges of state information R_Info, respectively. When the re-identification database 124 records more registered person feature vectors, the person identification system 100 will have a better ability to track people, but it will have a certain impact on the identification response speed.
[0035] The database comparison and update module 123 compares the person feature vector VC_IN with all registered person feature vectors in the re-identification database 124. If the difference (e.g., L2 distance (Euclidean distance)) between the registered person feature vector VCreg_k_1 and the person feature vector VC_IN found by the database comparison and update module 123 in the re-identification database 124 is less than a matching threshold THm, then it determines that the person feature vector VC_IN matches the person identity ID_k corresponding to the registered person feature vector VCreg_k_1. Therefore, the database comparison and update module 123 uses the person identity ID_k to identify the person CK in the captured image IMG_EXT. Furthermore, the database comparison and update module 123 determines which of the registered person feature vectors VCreg_k_1 to VCreg_k_3 related to person identity ID_k in the re-identification database 124 should be updated using the person feature vector VC_IN. Furthermore, the database comparison and update module 123 compares the state information R_Info (which may be the upper body proportion, the person's frontal rotation angle, the occlusion rate, the person's posture parameters, or a parameter that is a combination of the former) obtained by the state calculation module 122 from the captured image IMG_EXT with the update thresholds TH1 and TH2 (assuming that the update threshold TH2 is greater than the update threshold TH1) to determine the update method.
[0036] Specifically, if the state information R_Info of the captured image IMG_EXT is less than the update threshold TH1, the database comparison and update module 123 will use the person feature vector VC_IN to update the registered person feature vector VCreg_k_3 in the re-identification database 124 (assuming that the state information R_Info in the corresponding image is less than TH1); if the state information R_Info of the captured image IMG_EXT is greater than the update threshold TH1 but less than the update threshold TH2, the database comparison and update module 123 will use the person feature vector VC_IN to update the registered person feature vector VCreg_k_2 in the re-identification database 124 (assuming that the state information R_Info in the corresponding image is greater than TH1 but less than TH2); if the state information R_Info of the captured image IMG_EXT is greater than the update threshold TH2, the database comparison and update module 123 will use the person feature vector VC_IN to update the registered person feature vector VCreg_k_1 in the re-identification database 124 (assuming that the state information R_Info in the corresponding image is greater than TH2). Among them, the update thresholds TH1 and TH2 may vary depending on the specific person's state reflected by the state information R_Info.
[0037] Figure 4The update process of this embodiment of the invention is illustrated in this figure, which uses the state information R_Info as the upper body proportion for explanation. At time point t=N, the captured image IMG_EXT_1 containing the person CK is converted into a person feature vector VC_IN_1, and is compared by the database comparison and update module 123 with the registered person feature vectors in the re-identification database 124. Based on the L2 distance calculation, the person feature vector VC_IN_1 successfully matches the registered person feature vector corresponding to the person identity ID_k, so the person CK in the captured image IMG_EXT_1 is identified as person identity ID_k. Furthermore, according to the state information R_Info of the captured image IMG_EXT_1, its upper body proportion is less than the update threshold TH1 (i.e., the upper body proportion is 0-30%). Therefore, the person feature vector VC_IN_1 is used to update the registered person feature vector VCreg_k_3 in the re-identification database 124 (the upper body proportion of the image corresponding to which is 0-30%). Furthermore, at time point t = N+1, the captured image IMG_EXT_2 containing person CK is converted into a person feature vector VC_IN_2, which is then compared by the database comparison and update module 123 with the registered person feature vectors in the re-identification database 124. Based on the L2 distance calculation, the person feature vector VC_IN_2 successfully matches the registered person feature vector corresponding to person ID_k, so person CK in the captured image IMG_EXT_2 is identified as person ID_k. Moreover, according to the state information R_Info of the captured image IMG_EXT_2, its upper body proportion is greater than the update threshold TH2 (i.e., the upper body proportion is 60-90%). Therefore, the person feature vector VC_IN_2 is used to update the registered person feature vector VCreg_k_1 in the re-identification database 124 (the upper body proportion of the image corresponding to which is 60-90%). Through the above operations, it can be ensured that the re-identification database 124 always stores the feature vectors of registered persons corresponding to the upper body proportions of different numerical ranges.
[0038] On the other hand, if the database comparison and update module 123 does not find any registered person feature vector in the re-identification database 124 whose difference between the registered person feature vector and the person feature vector VC_IN is less than the matching threshold THm, then it is determined that the person feature vector VC_IN has not matched successfully. At this time, the database comparison and update module 123 will register a new person identity ID_w in the re-identification database 124, and based on the person feature vector VC_IN and the person feature vector obtained from the new image IMG_EXT in the subsequent time, register the registered vector related to the person identity ID_w in the re-identification database 124. Specifically, if the state information R_Info of the captured image IMG_EXT is less than the update threshold TH1, the database comparison and update module 123 registers the person feature vector VC_IN as the person feature vector VCreg_w_3 in the re-identification database 124; if the state information R_Info of the captured image IMG_EXT is greater than the update threshold TH1 but less than the update threshold TH2, the database comparison and update module 123 registers the person feature vector VC_IN as the person feature vector VCreg_w_2 in the re-identification database 124; if the state information R_Info of the captured image IMG_EXT is greater than the update threshold TH2, the database comparison and update module 123 registers the person feature vector VC_IN as the person feature vector VCreg_k_1 in the re-identification database 124. In some cases, if the status information R_Info in the subsequent captured image IMG_EXT continuously falls within the same numerical range, for example, if the status information R_Info is greater than the update threshold TH1 but less than the update threshold TH2, then the database comparison and update module 123 will not register the person feature vector VC_IN of the subsequent captured image IMG_EXT as person feature vectors VCreg_k_1 and VCreg_k_3, but will only continuously update the already registered person feature vector VCreg_k_2.
[0039] By employing the methods described above, it can be ensured that the re-identification database 124 covers as many feature vectors as possible for all state changes of the corresponding person. This allows the person identification system 100 to adapt better when the person's movements change drastically. Furthermore, if update thresholds TH1 and TH2 are not used to limit the updates of registered person feature vectors, the re-identification database 124 will always update the registered vectors with the successfully matched person feature vector VC_IN, which may have adverse effects. For example, if a person remains stationary for a long period (e.g., seated or unobstructed), continuous updates over a period will cause all registered person feature vectors in the re-identification database 124 to correspond to the person's seated or unobstructed state. Consequently, the person identification system 100 can only identify the person when they are seated or unobstructed. If the person suddenly undergoes a drastic change in state (e.g., from sitting to standing, or from unobstructed to obstructed), the person recognition system 100 will temporarily lose its ability to track and recognize the person because the re-identification database 124 lacks registered vectors related to the person's standing or obstructed state. Therefore, through the database registration and update management provided by this invention, it can be ensured that the person feature vector VC_IN with a specific state will only be used to update registered vectors with the same specific state in the re-identification database 124. For example, captured images showing a sitting or unobstructed state will only be used to update registered person feature vectors corresponding to the sitting or unobstructed state, and captured images showing a standing or obstructed state will only be used to update registered person feature vectors corresponding to the standing or obstructed state, thereby ensuring the diversity of updated registered person feature vectors in the re-identification database 124.
[0040] Figure 5 The figure illustrates a method for person re-identification in an embodiment of the present invention. As shown, the method of the present invention includes the following simplified process:
[0041] S310: Obtain a feature vector of a person based on a captured image containing a person;
[0042] S320: Based on the state of the person in the captured image, obtain state information;
[0043] S330: When the feature vector of a person is successfully matched with the first registered person feature vector among the multiple registered person feature vectors, the person is identified as the first person identity corresponding to the first registered person feature vector.
[0044] S340: When the feature vector of a person successfully matches a first registered person feature vector among the multiple registered person feature vectors, the person is identified as a first person identity corresponding to the first registered person feature vector; and
[0045] S350: Based on the status information, selectively use the person feature vector to update one of the first registered person feature vector and at least one second registered person feature vector corresponding to the first person identity.
[0046] Since the principles and specific details of the above steps have been described in detail in the previous embodiments, they will not be repeated here. It should be noted that the above process may be improved by adding other extra steps or making appropriate changes and adjustments to better achieve person re-identification and further enhance its recognition ability. Furthermore, all operations in the aforementioned embodiments of the present invention can be performed by… Figure 6 The device 400 shown is used to implement this. The storage unit 410 in the device 400 (which may be non-volatile memory, volatile memory, or a combination of both) can be used to store program code, instructions, variables, or data. The hardware processing unit (general-purpose processor) 420 in the device 400 can execute the program code and instructions stored in the storage unit 410, and refer to the variables or data therein to perform all the operations in the foregoing embodiments. In some embodiments of the present invention, one or more modules in the person re-identification system 100 can be implemented by a deep learning network (e.g., a convolutional neural network). Additionally, in some embodiments of the present invention, one or more modules in the person re-identification system 100 can also be implemented using pure hardware circuitry, such as an application-specific integrated circuit (ASIC), a programmable gate array (PGA), or a field-programmable gate array (FPGA).
[0047] In summary, this invention proposes a method and related apparatus for managing the updating and registration of a person re-identification database based on person state information, thereby enabling person re-identification. Specifically, to track changes in person state, this invention updates the person re-identification database in real time. Once a person feature vector calculated from the source image successfully matches a registered vector in the re-identification database, the invention updates the registered vector in the database using the current person feature vector. Furthermore, to ensure the re-identification database covers all possible changes in person state, this invention also references the current person state determined from the source image when registering and updating the database. The current person feature vector is only used to update registered vectors in the re-identification database that are similar to or related to the current person state. In this way, the person re-identification database can retain registered vectors that differ significantly from the current person state, ensuring accurate person identification even when the person state changes drastically.
[0048] Embodiments of the present invention can be implemented using hardware, software, firmware, and combinations thereof. Embodiments of the present invention can be implemented using software or firmware stored in memory and a corresponding instruction execution processor, provided a suitable instruction execution system is available. In terms of hardware, any of the following technologies or combinations thereof can be used: individual operational logic with logic gates capable of performing logical functions based on data signals; an application-specific integrated circuit (ASIC) with suitable combinational logic gates; a programmable gate array (PGA); or a field-programmable gate array (FPGA), etc.
[0049] The flowcharts and blocks in the specification illustrate the architecture, functionality, and operation achievable by systems, methods, and computer software products based on various embodiments of the invention. In this regard, each block in the flowchart or functional block diagram may represent a module, segment, or portion of program code, including one or more executable instructions for implementing a specified logical function. Furthermore, each block in the functional block diagram and / or flowchart, and combinations of blocks, can be implemented substantially by a dedicated hardware system that performs the specified function or action, or by a combination of dedicated hardware and computer program instructions. These computer program instructions may also be stored in a computer-readable medium that enables a computer or other programmable data processing apparatus to operate in a particular manner, such that the instructions stored in the computer-readable medium implement the function / action specified by the blocks in the flowchart and / or functional block diagram.
[0050] The above description is only a preferred embodiment of the present invention. All equivalent changes and modifications made in accordance with the claims of the present invention should be included within the scope of the present invention.
Claims
1. A method for re-identifying people, characterized in that, Include: Based on an image containing a person, a feature vector of that person is obtained; Based on the state of the person in the captured image, a state information is obtained; The feature vector of a person is matched with multiple registered person feature vectors in a single identification database. When the feature vector of a person successfully matches a first registered person feature vector among the multiple registered person feature vectors, the person is identified as the first person identity corresponding to the first registered person feature vector. Based on the status information, the first registered person feature vector corresponding to the first person identity is selectively updated using the person feature vector, and compared with one of at least one second registered person feature vector corresponding to the first person identity, wherein the status information is compared with at least one update threshold. When the status information falls within a numerical range defined by at least one update threshold, the first registered person's feature vector is updated using that person's feature vector; and When the status information does not fall within the numerical range, the feature vector of the at least one second registered person is updated using the person's feature vector.
2. The method as described in claim 1, characterized in that, The step of selectively using the person feature vector to update one of the first registered person feature vector and the at least one second registered person feature vector includes: Compare this status information with multiple update thresholds; When the status information falls within a first numerical range defined by the multiple update thresholds, the first registered person feature vector is updated using the person feature vector. When the status information falls within a second numerical range defined by the multiple update thresholds, the second registered person feature vector is updated using the person feature vector. as well as When the status information falls within a third numerical range defined by the multiple update thresholds, the third registered person feature vector corresponding to the first person's identity is updated using the person feature vector.
3. The method as described in claim 1, characterized in that, Also includes: When the feature vector of a person fails to match any of the multiple registered feature vectors of a person, a second person identity corresponding to that person is registered in the re-identification database. When the status information falls within a numerical range defined by at least one update threshold, the feature vector of the person is registered in the re-identification database as a fourth registered feature vector of the person corresponding to the second person's identity. as well as When the status information does not fall within the numerical range, the feature vector of the person is registered in the re-identification database as a fifth registered person feature vector corresponding to the identity of the second person.
4. The method as described in claim 1, characterized in that, This status information corresponds to one or more of the following parameters of the person in the captured image: the proportion of the person's upper body, the person's frontal rotation angle, the occlusion rate, and the person's pose.
5. A device for performing person re-identification, characterized in that, Include: A storage unit used to store a program code; A processing unit is used to execute the program code, so that the processing unit can perform the following operations: Based on an image containing a person, a feature vector of that person is obtained; Based on the state of the person in the captured image, a state information is obtained; The feature vector of this person is matched with the feature vectors of multiple registered persons in a single identification database; When the feature vector of a person successfully matches a first registered person feature vector among the multiple registered person feature vectors, the person is identified as the first person identity corresponding to the first registered person feature vector; and Based on the status information, the feature vector of the person is selectively used to update one of the first registered person feature vector corresponding to the first person identity, and at least one second registered person feature vector corresponding to the first person identity, wherein the status information is compared with at least one update threshold. When the status information falls within a numerical range defined by at least one update threshold, the first registered person's feature vector is updated using that person's feature vector; and When the status information does not fall within the specified range, use the character feature vector to update at least one second registered character feature vector.
6. The apparatus as claimed in claim 5, characterized in that, The processing unit executes the program code to perform the following operations: Compare this status information with multiple update thresholds; When the status information falls within a first numerical range defined by the multiple update thresholds, the first registered person feature vector is updated using the person feature vector. When the status information falls within a second numerical range defined by the multiple update thresholds, the second registered person feature vector is updated using the person feature vector. as well as When the status information falls within a third numerical range defined by the multiple update thresholds, the third registered person feature vector corresponding to the first person's identity is updated using the person feature vector.
7. The apparatus as claimed in claim 5, characterized in that, The processing unit executes the program code to perform the following operations: When the feature vector of a person fails to match any of the multiple registered feature vectors of a person, a second person identity corresponding to that person is registered in the re-identification database. When the status information falls within a numerical range defined by at least one update threshold, the feature vector of the person is registered in the re-identification database as a fourth registered feature vector of the person corresponding to the second person's identity. as well as When the status information does not fall within the numerical range, the feature vector of the person is registered in the re-identification database as a fifth registered person feature vector corresponding to the identity of the second person.
8. The apparatus as claimed in claim 5, characterized in that, This status information corresponds to one or more of the following parameters of the person in the captured image: the proportion of the person's upper body, the person's frontal rotation angle, the occlusion rate, and the person's pose.