A non-inductive check-in target tracking method and system based on trajectory credibility control
By using a trajectory credibility control method, a trajectory-level face feature cache structure is generated and trajectory credibility is calculated, which solves the problems of face feature contamination and identity recognition errors in complex scenarios and achieves stability and accuracy in target tracking and identity recognition.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- TIANJIN POLYTECHNIC UNIV
- Filing Date
- 2026-02-10
- Publication Date
- 2026-06-05
AI Technical Summary
Existing multi-target tracking technologies struggle to guarantee the accuracy and stability of attendance systems in complex, seamless check-in scenarios such as dense crowds, overlapping or occluded targets, due to miscorrelation of trajectories leading to facial feature contamination, identity recognition errors, and conflicts between trajectory continuity and identity consistency.
By using a trajectory credibility control method, a trajectory-level face feature cache structure is generated, trajectory credibility is calculated, and face feature updates are blocked when there is a conflict. Trajectory interruption or splitting is allowed, and temporal continuity is restored through trajectory posterior correlation to ensure identity consistency.
It effectively prevents facial feature contamination, ensures identity consistency, and improves the stability and reliability of target tracking and identity recognition systems in seamless check-in scenarios.
Smart Images
Figure CN122156252A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of computer vision and video analysis technology, specifically to a non-intrusive check-in target tracking method and system based on trajectory credibility control. Background Technology
[0002] In the field of multi-object tracking (MOT), existing mainstream methods typically employ a "detection-tracking" framework. This framework associates the trajectory based on the spatial position changes of the target in adjacent video frames and motion prediction models (such as Kalman filtering), and maintains the temporal continuity of the trajectory through matching strategies such as the Hungarian algorithm. While these methods achieve good tracking results in simple scenarios, their design prioritizes trajectory continuity and lacks effective constraints on maintaining the long-term consistency of the target's identity. To mitigate the identity switching problem, some methods introduce re-identification (ReID) features. For example, in DeepSORT and its improved methods, as well as some fusion algorithms based on the ByteTrack framework, appearance features or facial features are typically combined with the target's motion information to assist in trajectory association and identity determination. However, these methods usually treat ReID features as continuously participating enhancement information in the matching process and update the feature model according to predetermined rules after the trajectory association is established, lacking an effective control mechanism for the reliability of feature writing.
[0003] In complex scenarios involving multiple people moving in parallel, intersecting, or with occlusion, the trajectory association results may still be uncertain due to short-term occlusion and unstable detection results. Incorrect features written in the early stages under conditions of insufficient association credibility may be retained for a long time and continue to participate in subsequent decisions, causing identity recognition to be interfered with by accumulated errors. Especially in application scenarios such as contactless check-in, where personnel passage space is narrow and interactions are frequent, the conflict between trajectory continuity and identity consistency is more prominent. Existing methods are unable to proactively constrain feature update behavior or adjust trajectory structure based on the credibility of association results, thus failing to meet the application requirements of identity recognition accuracy and system stability in contactless check-in scenarios. Summary of the Invention
[0004] To address the aforementioned issues, this invention proposes a seamless check-in target tracking method based on trajectory credibility control. In the target tracking process, facial features are not simply used as enhancement features for continuous matching. Instead, trajectory credibility is used to control the updating behavior of facial features and the continuity constraints of the trajectory.
[0005] The present invention also provides a seamless check-in system for executing the above-mentioned target tracking method based on trajectory credibility control. The system coordinates and controls the generation of target tracking trajectory, the maintenance of facial feature cache, and trajectory correction processing through the collaborative work between various functional modules, thereby improving identity consistency and system stability in complex scenarios.
[0006] The purpose of this invention is to: actively block facial feature updates and allow trajectory interruption or splitting when there is a conflict between the continuity of target motion and the consistency of facial features, thereby preventing facial feature contamination; and perform posterior association or merging of trajectory segments based on trajectory credibility and facial feature consistency in subsequent video frames, thereby restoring the temporal continuity of the trajectory while ensuring identity consistency.
[0007] The technical problem to be solved by this invention is that existing multi-target tracking technologies have problems such as facial feature contamination, identity recognition errors, and conflicts between trajectory continuity and identity consistency in complex and non-intrusive check-in scenarios such as dense crowds, overlapping or occluded targets. These problems make it difficult to ensure the accuracy and stability of the attendance system.
[0008] The present invention solves the above problems through the following technical solution, including the following steps:
[0009] S1, Trajectory Generation Steps: The acquired video sequence is processed frame by frame. The detection results of human targets are obtained in each video frame. Based on the spatial position relationship of human targets in adjacent video frames and the motion prediction model, a corresponding target tracking trajectory is generated for each human target.
[0010] S2, Trajectory-level face feature cache maintenance step: Maintain a corresponding trajectory-level face feature cache structure for each target tracking trajectory, which is used to store multi-frame face feature vectors associated with the target tracking trajectory, and generate face representative features based on the multi-frame face feature vectors to characterize the human identity corresponding to the target tracking trajectory.
[0011] S3, Trajectory credibility calculation steps: When a face image corresponding to the target tracking trajectory is detected in the current video frame, the face feature vector corresponding to the face image in the current video frame is extracted, and the trajectory credibility of the target tracking trajectory in the current video frame is calculated based at least on the target's motion prediction deviation and the feature similarity between the face feature vector and the face representative features.
[0012] S4, Face Feature Writing Control Step: Based on whether the trajectory confidence and the current face feature vector meet the preset confidence conditions, control whether to allow the face feature vector extracted from the current video frame to be written into the trajectory-level face feature cache structure; when the trajectory confidence and the face feature vector extracted from the current video frame do not meet the preset confidence conditions, block the writing of face features to the trajectory-level face feature cache structure to prevent feature pollution caused by the writing of erroneous face features;
[0013] S5, Trajectory Conflict Handling Steps: When the trajectory judgment result obtained based on the continuity of target motion conflicts with the face feature consistency judgment result obtained based on the trajectory-level face feature cache structure, the face feature vectors already stored in the trajectory-level face feature cache structure are not updated, and the constraints on the continuity of the target tracking trajectory are relaxed, allowing the target tracking trajectory to be interrupted or split, so as to generate multiple trajectory segments.
[0014] S6, Trajectory Post-Association Step: In subsequent video frames, the trajectory segments are post-associated or merged only when the trajectory credibility of at least one trajectory segment in the candidate trajectory segments meets the trajectory credibility-related part of the preset credibility condition, and the facial representation features between the corresponding trajectory segments meet the consistency condition, so as to generate the corrected target tracking trajectory.
[0015] Furthermore, in step S2, the representative facial features are not directly determined by single-frame facial features, but are generated by weighting the facial feature vectors of multiple frames that meet the writing conditions in the trajectory-level facial feature cache structure according to the trajectory confidence of the corresponding frames. The representative facial features can be calculated according to the method described in equation (1-1):
[0016]
[0017] Among them, f rep f represents facial features i Let C represent the facial feature vector corresponding to the i-th frame. i This indicates the trajectory reliability corresponding to the frame, and K represents the number of facial feature vectors generated.
[0018] Furthermore, in step S3, the reliability of the target tracking trajectory in the current video frame can be calculated based at least on the target's motion prediction deviation and the consistency of facial features. The motion prediction deviation of the human target can be represented by equation (1-2):
[0019]
[0020] Where, d t This indicates the motion prediction error of the human target in the current video frame. This indicates the detected position of the human target in the current video frame. This indicates the predicted location obtained through a motion prediction model based on historical trajectories.
[0021] Furthermore, in step S3, the feature similarity between the face feature vector extracted from the current video frame and the trajectory-level face representative features can be represented by equation (1-3):
[0022]
[0023] Among them, s t f represents the feature similarity between the face feature vector extracted from the current video frame and the trajectory-level face representation features. t f represents the facial feature vector extracted from the current video frame. rep This represents the facial representation features generated by the trajectory-level facial feature cache structure.
[0024] Furthermore, in step S3, the trajectory reliability is used to characterize the reliability of the current target tracking trajectory in the current video frame. The trajectory reliability is not used to select the optimal trajectory matching result, but rather as a control variable to determine whether to allow facial feature writing, whether to relax trajectory continuity constraints, and whether to perform trajectory posterior association. The trajectory reliability can be comprehensively calculated according to the method described in equations (1-4):
[0025] C t =α·g(d t )+β·s t (1-4)
[0026] Among them, C t Let α and β be the trajectory confidence level, and let α+β=1. g(·) is a mapping function for motion prediction bias, which is used to convert motion bias into a confidence metric comparable to feature similarity.
[0027] Furthermore, in step S4, the preset credibility conditions include at least: the trajectory credibility is higher than a preset credibility threshold; the trajectory credibility remains stable within at least one preset number of consecutive frames; the feature similarity between the current face feature vector and the face representative feature is higher than a preset consistency threshold, and the stability of the trajectory credibility can be determined by the conditions described in equations (1-5):
[0028] |C t -C t-1 |≤δ,t=1,2,...,N (1-5)
[0029] Where δ is the preset confidence fluctuation threshold and N is the number of consecutive video frames, used to characterize the stability requirements of trajectory confidence in the time dimension.
[0030] The present invention also provides a seamless check-in system based on the above-mentioned target tracking method, comprising:
[0031] Personnel Basic Information Management Module: Used to store, maintain, and manage personnel's basic identity information and corresponding facial feature data, and to provide the facial recognition module with the reference information required for personnel identity comparison;
[0032] Image acquisition module: used to acquire video image data of the attendance area, and when a scene change or acquisition trigger event is detected in the attendance area, output the corresponding video image sequence or image frame to the human body detection module;
[0033] Human detection module: used to detect human targets in video images provided by the image acquisition module;
[0034] Human body tracking module: used to generate target tracking trajectory based on human body detection results, maintain trajectory-level face feature cache structure and calculate trajectory credibility; when the trajectory judgment result obtained based on the continuity of target motion conflicts with the judgment result obtained based on the consistency of face features, the constraint on the continuity of the target tracking trajectory is relaxed, allowing the target tracking trajectory to be interrupted or split in the current video frame to generate multiple trajectory segments;
[0035] The trajectory correction module is used to perform posterior association or merging processing on the candidate trajectory segments generated by the human body tracking module after the video sequence processing is completed, based on the trajectory credibility and facial representation feature consistency of each candidate trajectory segment generated by the human body tracking module, to generate a corrected target tracking trajectory.
[0036] Face recognition module: Used to compare the facial features corresponding to the corrected target tracking trajectory with the reference features provided by the personnel basic information management module to achieve identity recognition.
[0037] The modules described above can be implemented using hardware, software, or a combination of both, and their functional configurations are used to execute the various steps of the aforementioned target tracking method.
[0038] Compared with the prior art, the present invention has at least the following beneficial effects:
[0039] 1. By controlling facial feature updates based on trajectory reliability, the problem of facial feature contamination caused by erroneous trajectory association can be effectively prevented;
[0040] 2. When a conflict arises between motion continuity and facial feature consistency, the target tracking trajectory can be interrupted or split to ensure identity verification at the system level.
[0041] consistency;
[0042] 3. By using a trajectory post-hoc association mechanism, the temporal continuity of the trajectory is restored while ensuring identity consistency;
[0043] 4. Improved the stability and reliability of target tracking and identity recognition systems in complex scenarios such as contactless check-in. Attached Figure Description
[0044] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0045] Figure 1 This is a flowchart illustrating the non-intrusive check-in target tracking method based on trajectory reliability control in an embodiment of the present invention.
[0046] Figure 2 This is a schematic diagram of the trajectory reliability control logic in an embodiment of the present invention;
[0047] Figure 3 This is a system module structure diagram in an embodiment of the present invention. Detailed Implementation
[0048] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.
[0049] This implementation is based on the following typical application scenario: The camera is a monocular RGB camera installed at a height of approximately 1.7 meters, and the shooting position is located at the entrance of the office area. Since the area is equipped with a glass door, the camera can simultaneously cover part of the area inside and outside the door. During peak hours, people frequently enter and exit, and complex target tracking situations such as multiple people walking side by side, crossing each other, and short-term obstruction often occur.
[0050] The embodiments of the present invention further illustrate the effect of the above processing flow with the following example scenario: In a video sequence containing multiple people, two people enter an office area in parallel or intersecting directions. In the intersecting area, although the target tracking trajectory based on motion prediction remains continuous in space, the ID identifier is incorrectly updated due to occlusion between people. Initially, the trajectory number of person A is ID1, and the trajectory number of person B is ID2. When the two pass through the entrance at the same time and occlude in the doorway area, the existing multi-target tracking method incorrectly assigns person A to ID2 due to recognition confusion, while person B is misidentified as a new target ID3, causing the original ID1 trajectory to be interrupted and the identity identifier to be incorrectly updated.
[0051] This invention provides a seamless check-in target tracking method based on trajectory reliability control. The following description, in conjunction with specific application scenarios, further illustrates the effectiveness of this method. Figure 1 As shown, the specific steps are as follows:
[0052] S1, process the acquired video sequence frame by frame and obtain the detection results of human targets, and generate corresponding target tracking trajectories for two people based on the spatial position relationship of human targets in adjacent video frames and motion prediction models;
[0053] S2, write the multi-frame facial feature vectors associated with the trajectories of the two people into the corresponding trajectory-level facial feature cache structure, and generate representative facial features of the two people based on the multi-frame facial feature vectors.
[0054] S3, extract the face images of the two people in the current video frame respectively, and calculate the trajectory credibility of the two people in the current video frame based on the target motion prediction deviation and the face feature similarity.
[0055] S4. Based on whether the trajectory credibility and the current face feature vector meet the preset credibility conditions, control whether to allow the face feature vector extracted from the current video frame to be written into the trajectory-level face feature cache structure. Only when the trajectory credibility and the face feature vector extracted from the current video frame meet the preset credibility conditions is it allowed to write the face feature vector extracted from the current video frame into the trajectory-level face feature cache structure to prevent feature pollution caused by incorrect face features being written.
[0056] S5, when the trajectory judgment result obtained based on the continuity of target motion conflicts with the face feature consistency judgment result obtained based on the trajectory-level face feature cache structure, the corresponding trajectory-level face feature cache structure is frozen, the constraint on the continuity of target tracking trajectory is relaxed, and the target tracking trajectory is allowed to be interrupted or split to generate multiple trajectory segments.
[0057] S6, in subsequent video frames, the trajectory segments are only associated or merged posteriorly to generate a corrected target tracking trajectory if the trajectory credibility of at least one trajectory segment in the candidate trajectory segments satisfies the trajectory credibility-related part of the preset credibility condition, and the facial representation features between the corresponding trajectory segments satisfy the consistency condition.
[0058] Furthermore, in step S3, after a parallel or cross event occurs, person A is mistakenly associated with person B. The corresponding ID2 trajectory remains continuous in space, but the feature similarity between the face feature vector extracted in the current video frame and the trajectory-level face representative features is significantly reduced, resulting in a significant decrease in the reliability of the ID2 trajectory.
[0059] Furthermore, in step S4, after a parallel or cross event occurs, the credibility of the ID2 trajectory is significantly reduced, and the preset credibility condition is not met, thus blocking the writing of the current person A's facial features to the facial feature cache structure of the ID2 trajectory.
[0060] Furthermore, in step S5, after a parallel or cross event occurs, the trajectory judgment result obtained by the ID2 trajectory based on the continuity of the target motion conflicts with the face feature consistency judgment result obtained based on the trajectory-level face feature cache structure. The ID2 trajectory is automatically split into two trajectories, ID2 and ID4, in the current frame.
[0061] Furthermore, in step S6, after the video sequence processing is completed, a total of four candidate trajectory segments are obtained. Two of these segments meet the requirements of trajectory credibility and consistency of facial representation features. The segments are then correlated or merged posteriorly to generate two corrected target tracking trajectories. The corrected target tracking trajectories do not exhibit facial feature contamination.
[0062] In this embodiment, trajectory confidence is not used to select the optimal trajectory matching result, but rather as a control variable to determine whether facial features are written, whether trajectory continuity constraints are relaxed, and whether trajectory posterior association is performed. Figure 2 As shown, the control logic for trajectory reliability includes the following process:
[0063] S1, Trajectory credibility calculation process: In the current video frame, when a face image corresponding to the target tracking trajectory is detected, the trajectory credibility of the target tracking trajectory in the current video frame is comprehensively calculated based on the motion prediction deviation between the target's motion prediction position and the actual detection position, as well as the feature similarity between the face feature vector extracted in the current video frame and the trajectory-level face representative features.
[0064] S2, Face Feature Writing Judgment and Control Process: The calculated trajectory credibility is judged against the preset credibility conditions. When the trajectory credibility is higher than the preset credibility threshold and remains stable within at least one preset number of consecutive frames, and the feature similarity between the current face feature vector and the trajectory-level face representative feature meets the consistency requirements, the face feature vector extracted from the current video frame is allowed to be written into the corresponding trajectory-level face feature cache structure. When the trajectory credibility or face feature consistency does not meet the preset credibility conditions, the writing of the current face feature to the trajectory-level face feature cache structure is blocked to prevent incorrect face features from being written and causing feature contamination.
[0065] S3, Trajectory Continuity Constraint Adjustment Process: When the trajectory judgment result obtained based on the continuity of target motion conflicts with the face feature consistency judgment result obtained based on the trajectory-level face feature cache structure, the face feature vectors already stored in the trajectory-level face feature cache structure are not updated, and the constraint on the continuity of target tracking trajectory is relaxed, allowing the target tracking trajectory to be interrupted or split in the current video frame to generate multiple trajectory segments.
[0066] S4, Trajectory Post-Association and Merging Control Process: In subsequent video frames, multiple trajectory segments generated by trajectory interruption or split are evaluated. Only when the trajectory credibility of at least one trajectory segment among the candidate trajectory segments meets the requirements related to trajectory credibility in the preset credibility conditions, and the facial representation features between the corresponding trajectory segments meet the consistency conditions, is the post-association or merging between trajectory segments performed to generate the corrected target tracking trajectory.
[0067] This invention also provides a seamless check-in system based on trajectory reliability control, used to run the aforementioned seamless check-in target tracking method based on trajectory reliability control, such as... Figure 3 As shown, it includes:
[0068] (1) Personnel Basic Information Management Module: Used to store, maintain and manage the basic identity information and corresponding facial feature data of the personnel to be identified, and to provide the reference facial feature vector required for target comparison to the facial recognition module, so as to realize the feature database support required for identity confirmation;
[0069] (2) Image acquisition module: used to acquire video image data of the attendance area, and when a scene change or acquisition trigger event is detected in the attendance area, output the corresponding video image sequence or image frame to the human body detection module for subsequent human body detection and target tracking processing.
[0070] (3) Human detection module: used to process the video images provided by the image acquisition module, detect and output the position information of human targets in each frame of the image, and use them to generate tracking trajectories in the subsequent target tracking module;
[0071] (4) Human tracking module: used to generate target tracking trajectory based on the detection results output by the human detection module, and maintain a corresponding trajectory-level face feature cache structure for each target tracking trajectory; when processing video sequences frame by frame, the trajectory credibility is calculated based on the target motion prediction deviation and face feature consistency, and the face features are controlled to be written into the trajectory-level face feature cache structure according to the trajectory credibility; when the trajectory judgment result obtained based on the target motion continuity conflicts with the judgment result obtained based on the face feature consistency, the target tracking trajectory is allowed to be interrupted or split to generate multiple trajectory segments;
[0072] (5) Trajectory correction module: After the video sequence processing is completed, multiple candidate trajectory segments generated by the human body tracking module are obtained; based on the trajectory credibility and face representation feature consistency of each candidate trajectory segment, the candidate trajectory segments are subjected to posterior association or merging processing to generate the corrected target tracking trajectory.
[0073] (6) Face recognition module: used to extract the facial representation features of the corrected target tracking trajectory and match them with the reference features provided by the personnel basic information management module to obtain the identity information of the target personnel and realize identity recognition.
[0074] In summary, the contactless check-in target tracking method and system based on trajectory credibility control described in the above embodiments, targeting application scenarios that are prone to target identity confusion such as dense crowds, target intersections, and occlusions, achieves a target tracking strategy that prioritizes face consistency by controlling facial feature updates based on trajectory credibility, relaxing the continuity constraints of target tracking trajectories when necessary, and performing trajectory posterior association processing. This improves the stability and reliability of target tracking and identity recognition systems in complex scenarios such as contactless check-in, and is worthy of widespread adoption.
[0075] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention are included within the scope of protection of the present invention.
Claims
1. A method and system for seamless check-in target tracking based on trajectory reliability control, characterized in that, Includes the following steps: S1, the video sequence is processed frame by frame, the detection result of human target is obtained in each video frame, and the corresponding target tracking trajectory is generated for each human target based on the spatial position relationship of human target in adjacent video frames and motion prediction model. S2, maintain a corresponding trajectory-level face feature cache structure for each target tracking trajectory. The trajectory-level face feature cache structure is used to store multiple frames of face feature vectors associated with the target tracking trajectory, and generate face representative features to characterize the human identity corresponding to the target tracking trajectory based on the multiple frames of face feature vectors. S3, when a face image corresponding to the target tracking trajectory is detected in the current video frame, the face feature vector corresponding to the face image in the current video frame is extracted, and the trajectory credibility of the target tracking trajectory in the current video frame is calculated based at least on the motion deviation between the predicted motion position and the actual detection position of the target in the current video frame and the feature similarity between the current face feature vector and the face representative feature. The trajectory credibility is used to characterize the reliability of the current target tracking trajectory in the current video frame. The trajectory credibility is not used to select the optimal trajectory matching result, but is used as a control variable to determine whether to allow face feature writing, whether to relax the trajectory continuity constraint, and whether to perform trajectory posterior association. S4, based on whether the trajectory confidence and the current face feature vector meet the preset confidence conditions, control whether to allow the face feature vector extracted from the current video frame to be written into the trajectory-level face feature cache structure; when the trajectory confidence and the face feature vector extracted from the current video frame do not meet the preset confidence conditions, block the writing of face features to the trajectory-level face feature cache structure to prevent feature pollution caused by the writing of erroneous face features; S5, when the trajectory judgment result obtained based on the continuity of target motion conflicts with the face feature consistency judgment result obtained based on the trajectory-level face feature cache structure, the face feature vector stored in the trajectory-level face feature cache structure is not updated and the constraint on the continuity of the target tracking trajectory is relaxed, allowing the target tracking trajectory to be interrupted or split in the current video frame to generate multiple trajectory segments; S6, in subsequent video frames, the trajectory segments are only associated or merged posteriorly to generate a corrected target tracking trajectory if the trajectory credibility of at least one trajectory segment in the candidate trajectory segments satisfies the trajectory credibility-related part of the preset credibility condition, and the facial representation features between the corresponding trajectory segments satisfy the consistency condition. In step S2, the representative facial features are not directly determined by single-frame facial features, but are generated by weighting the facial feature vectors of multiple frames that meet the writing conditions in the trajectory-level facial feature cache structure according to the trajectory confidence of the corresponding frames. The representative facial features can be calculated according to the method described in equation (1-1): Among them, f rep f represents facial features i Let C represent the facial feature vector corresponding to the i-th frame. i This indicates the reliability of the trajectory corresponding to the frame, and K represents the number of facial feature vectors generated. In step S3, the reliability of the target tracking trajectory in the current video frame can be calculated based at least on the target's motion prediction deviation and the consistency of facial features. The motion prediction deviation of the human target can be represented by equation (1-2): Where, d t This indicates the motion prediction error of the human target in the current video frame. This indicates the detected position of the human target in the current video frame. This represents the predicted location obtained through a motion prediction model based on historical trajectories. In step S3, the feature similarity between the face feature vector extracted from the current video frame and the trajectory-level face representative features can be represented by equation (1-3): Among them, s t f represents the feature similarity between the face feature vector extracted from the current video frame and the trajectory-level face representation features. t f represents the facial feature vector extracted from the current video frame. rep This represents the facial representation features generated by the trajectory-level facial feature cache structure. In step S3, the reliability of the trajectory can be calculated comprehensively according to the method described in equations (1-4): C t =α·g(d t )+β·s t (1-4) Among them, C t Let α and β be the trajectory confidence level, and let α + β = 1. g(·) is a mapping function for motion prediction bias, which is used to convert motion bias into a confidence metric comparable to feature similarity. In step S4, the preset credibility conditions include at least: the trajectory credibility is higher than a preset credibility threshold; the trajectory credibility remains stable within at least one preset number of consecutive frames; the feature similarity between the current face feature vector and the face representative feature is higher than a preset consistency threshold, and the stability of the trajectory credibility can be judged by the conditions described in equation (1-5): |C t -C t-1 |≤δ,t=1,2,...,N (1-5) Where δ is the preset confidence fluctuation threshold and N is the number of consecutive video frames, used to characterize the stability requirements of trajectory confidence in the time dimension.
2. The target tracking method according to claim 1, characterized in that: The trajectory-level face feature caching structure only allows the writing of the face feature vector extracted from the current video frame when the target tracking trajectory is in an unobstructed or low-obstructed state.
3. The target tracking method according to claim 2, characterized in that: The determination of the unobstructed or low-obstruction state is based at least on the overlap between human targets and the spatial inclusion relationship between the human detection box and the face detection box.
4. The target tracking method according to claim 1, characterized in that: The target tracking method is applied to the contactless check-in scenario at the entrance and exit of the office area. When the width of the personnel entry and exit channel is less than a preset threshold, and at least two people enter and exit at the same time within the same time window, a joint constraint mechanism based on trajectory credibility and facial feature consistency is activated.
5. A seamless check-in system, used to execute the seamless check-in target tracking method based on trajectory reliability control as described in claim 1, characterized in that, include: The system includes modules for basic personnel information management, image acquisition, human body detection, human body tracking, trajectory correction, and facial recognition. The human tracking module is configured to: generate a target tracking trajectory based on the detection results output by the human detection module, and maintain a corresponding trajectory-level face feature cache structure for each target tracking trajectory; when processing the video sequence frame by frame, calculate the trajectory credibility based on the target's motion prediction deviation and face feature consistency, and control whether face features are written into the trajectory-level face feature cache structure based on the trajectory credibility; when the trajectory judgment result obtained based on the target motion continuity conflicts with the judgment result obtained based on face feature consistency, relax the constraint on the continuity of the target tracking trajectory, and allow the target tracking trajectory to be interrupted or split in the current video frame to generate multiple trajectory segments; The trajectory correction module is configured to: after the video sequence processing is completed, acquire multiple candidate trajectory segments generated by the human body tracking module; based on the trajectory credibility and facial representation feature consistency of each candidate trajectory segment, perform posterior association or merging processing on the candidate trajectory segments to generate the corrected target tracking trajectory.