Personnel tracking method and system based on authenticity and occlusion recovery

Abstract

The invention discloses a personnel tracking method and system based on authenticity and shielding recovery, belongs to the technical field of personnel tracking, and remarkably improves the tracking accuracy and continuity by combining authenticity evaluation and shielding recovery technologies aiming at the shielding problem in a complex scene. The method comprises the following steps: firstly identifying tracking personnel in a video frame, and then evaluating whether a region of interest is shielded or not; for the occlusion area, authenticity evaluation is carried out according to appearance characteristics, and if the authenticity of evaluation is high, the position after occlusion elimination is predicted and the track is updated; and if not, re-detecting and recovering the track immediately. And if the authenticity of the evaluation is low, shielding recovery and re-evaluation are carried out, so that the accuracy of target locking is ensured. The method not only effectively deals with shielding challenges, but also ensures that the tracking trajectory is continuous and errorless through an intelligent re-detection and trajectory recovery mechanism, and provides powerful support for efficient personnel tracking in a complex environment.

Description

Technical Field

[0001] The present invention relates to the technical field of personnel tracking, and more particularly to a personnel tracking method and system based on authenticity and occlusion recovery. Background Art

[0002] With the rapid development of computer vision technology, people tracking technology, as one of its core applications, has shown great application potential and value in many fields such as security monitoring, behavior analysis, and intelligent transportation. However, although individual human tracking technology has made significant progress and has reached a practical stage in terms of tracking accuracy and speed, people tracking technology still faces many challenges, the most prominent of which is the occlusion problem.

[0003] In complex scenes, the tracked target often experiences occlusion, including self-occlusion, pedestrian occlusion, and background occlusion. These occlusions can cause significant changes in the appearance of the tracked target, making it difficult for the tracking algorithm to accurately identify and track the target, affecting the stability of the tracking algorithm, and ultimately leading to the loss of the tracked person or incomplete tracking trajectory.

[0004] At present, when dealing with the occlusion problem in multi-body tracking, most methods are used to predict the target state by trajectory matching or establishing motion equations. These methods can achieve good results when the target motion trajectory is relatively simple, but in most cases, due to the uncertainty of the target's motion direction and speed, the trajectory and motion state are difficult to accurately predict, so these methods have obvious limitations in solving the occlusion problem. Trajectory matching relies on the matching of historical trajectory data, but in the case of occlusion, the target trajectory is often discontinuous, resulting in matching failure. Motion equation prediction predicts the future state of the target by establishing a motion model of the target, but in the case of occlusion, the target's motion state may mutate, making the prediction result inaccurate.

[0005] Therefore, how to provide a personnel tracking method and system that can improve the accuracy of personnel tracking trajectory is a problem that technical personnel in this field urgently need to solve. Summary of the invention

[0006] In view of this, the present invention provides a method and system for tracking people based on authenticity and occlusion recovery. Combining authenticity and occlusion recovery not only improves the accuracy of tracking objects, but also improves the accuracy and integrity of tracking trajectories.

[0007] In order to achieve the above object, the present invention adopts the following technical solution:

[0008] On the one hand, the present invention provides a method for tracking people based on authenticity and occlusion recovery, comprising:

[0009] Step 1: Obtain the video frame image to be detected;

[0010] Step 2: Using a personnel recognition model to recognize the video frame image to be detected, and obtaining a region of interest of the tracked person;

[0011] Step 3: Determine whether the region of interest is blocked. If not, obtain the tracking result and update the tracking trajectory. If there is blocking, perform authenticity evaluation on the region of interest based on the appearance features of the region of interest and the tracked person, and select different processing methods based on the authenticity evaluation results.

[0012] Step 4: If the authenticity evaluation result is greater than the preset condition, the trajectory is predicted based on the tracking trajectory and the position of the tracked person, the predicted position and predicted time of the tracked person after the occlusion is eliminated are obtained, and the tracked person is matched at the predicted time and the predicted position. If the tracked person is matched, the tracking trajectory is updated based on the predicted trajectory; if the tracked person is not matched, the current position of the tracked person is obtained, and the tracking trajectory is restored based on the current position of the tracked person and the occlusion position;

[0013] Step 5: If the authenticity evaluation result is less than or equal to the preset condition, perform occlusion recovery on the region of interest, and evaluate the authenticity of the region of interest after occlusion recovery based on the occlusion recovery result; if the authenticity evaluation result of the region of interest after occlusion recovery is greater than the preset condition, return to step 4; if the authenticity evaluation result of the region of interest after occlusion recovery is less than or equal to the preset condition, obtain the current position of the tracked person, and restore the tracking trajectory based on the current position of the tracked person and the occlusion position.

[0014] Preferably, determining whether the region of interest is blocked specifically includes:

[0015] Divide multiple appearance feature key point areas according to the appearance features of the tracked person;

[0016] Matching the appearance feature key point region within the region of interest;

[0017] The number of the appearance feature key point regions matched in the region of interest is counted, and if the number is less than the number of the appearance feature key point regions, it is determined that the region of interest is blocked.

[0018] Preferably, performing authenticity assessment on the region of interest based on appearance features of the region of interest and the tracked person includes:

[0019] Calculating the occlusion determination values ​​of the appearance feature key point areas respectively;

[0020] Determining the importance index of the appearance feature key point area based on the appearance feature;

[0021] The occlusion determination value is weighted according to the importance index, and a weighted sum is performed, and the authenticity evaluation result of the region of interest is obtained according to the weighted sum result.

[0022] Preferably, calculating the occlusion determination values ​​of the appearance feature key point areas respectively includes:

[0023] For an unmatched appearance feature key point region, first appearance information of the unmatched appearance feature key point region and second appearance information of the appearance feature key point region corresponding to the region of interest are respectively obtained; based on the first appearance information, the second appearance information and an occlusion determination function, an occlusion determination value corresponding to the appearance feature key point region is determined;

[0024] For the matched appearance feature key point area, the occlusion determination value is determined to be zero.

[0025] Preferably, performing trajectory prediction based on the tracking trajectory and the position of the tracked person, and obtaining the predicted position and predicted time of the tracked person after the occlusion is eliminated, includes:

[0026] Obtain historical trajectory point information of the occluder, and determine whether the occluder is a static target based on the historical trajectory point information;

[0027] If the occluder is a static target, the turning probability of the tracked person is predicted based on the position information of the occluder and the position of the tracked person, and the uniform speed model of the tracked person is determined based on the tracking trajectory, and the predicted position and predicted time of the tracked person after the occlusion is eliminated are predicted based on the position of the tracked person, the uniform speed model of the tracked person, and the turning probability;

[0028] If the obstruction is a dynamic target, the turning probability of the tracked person and the turning probability of the obstruction are predicted respectively based on the position information of the obstruction and the position of the tracked person, and the uniform speed model of the tracked person is determined based on the tracking trajectory, and the uniform speed model of the obstruction is determined according to the historical trajectory point information. Based on the position of the tracked person, the uniform speed model of the tracked person, the position of the obstruction, the uniform speed model of the obstruction, the turning probability of the tracked person, and the turning probability of the obstruction, the predicted position and predicted time of the tracked person after the occlusion is eliminated are predicted.

[0029] Preferably, performing occlusion recovery on the region of interest includes:

[0030] Performing semantic segmentation on the region of interest using a semantic segmentation model, removing occluders, and obtaining an occluder region;

[0031] Reducing the dimension of the occluder region, and converting the RGB color space of the occluder region into a YCbCr color space;

[0032] Determine a contour point set of the occluder area, repair each pixel in the contour point set, and update the contour point set, removing the repaired pixel points until the contour point set is empty;

[0033] After completing the repair of all pixels, the repaired area is obtained;

[0034] Convert the YCbCr color space of the repaired area back to RGB color space.

[0035] On the other hand, the present invention further provides a system for implementing the above-mentioned person tracking method based on authenticity and occlusion recovery, comprising:

[0036] An acquisition module, used for acquiring the video frame image to be detected;

[0037] A region extraction model is used to identify the video frame image to be detected using a person recognition model to obtain a region of interest for tracking the person;

[0038] An occlusion judgment module is used to judge whether there is occlusion in the region of interest;

[0039] An evaluation module, for evaluating the authenticity of the region of interest according to the appearance features of the region of interest and the tracked person if there is occlusion, and determining to select different processing methods according to the authenticity evaluation result;

[0040] An update module, used to obtain the tracking result and update the tracking trajectory if there is no occlusion;

[0041] A personnel matching module is used to predict the track based on the tracking track and the position of the tracked person if the authenticity evaluation result is greater than a preset condition, obtain the predicted position and predicted time of the tracked person after the occlusion is eliminated, and match the tracked person at the predicted time and the predicted position. If the tracked person is matched, the tracking track is updated based on the predicted track;

[0042] If the tracked person is not matched, the current position of the tracked person is obtained, and the tracking trajectory is restored based on the current position of the tracked person and the occlusion position;

[0043] An occlusion recovery module is used to perform occlusion recovery on the region of interest if the authenticity evaluation result is less than or equal to a preset condition, evaluate the authenticity of the region of interest after occlusion recovery based on the occlusion recovery result, and return to the person matching module if the region of interest after occlusion recovery is greater than a preset value;

[0044] The recovery tracking module obtains the current position of the tracked person if the tracked person is not matched or the area of ​​interest after the occlusion is restored is less than or equal to a preset condition, and recovers the tracking trajectory based on the current position of the tracked person and the occlusion position.

[0045] It can be seen from the above technical solutions that compared with the prior art, the present invention discloses a personnel tracking method and system based on authenticity and occlusion recovery. First, the authenticity of the occluded area is evaluated, and the authenticity of the identity of the tracked object in a complex environment can be evaluated to ensure continuous and accurate locking of the correct tracking target. Secondly, the present invention also improves the accuracy of locking the tracked person in the case of occlusion by performing occlusion recovery on the occluded area with low authenticity. Furthermore, for the tracked person who is not successfully matched after the occlusion is eliminated, the present application uses an efficient re-detection mechanism to quickly locate the target, and combines the motion model to perform re-trajectory recovery to ensure the continuity and accuracy of the tracking trajectory. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0047] Figure 1 A schematic diagram of a flow chart of a personnel tracking method provided by the present invention;

[0048] Figure 2 A framework diagram of the personnel tracking system provided by the present invention. DETAILED DESCRIPTION

[0049] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0050] The embodiment of the present invention discloses a method for tracking people based on authenticity and occlusion recovery. Figure 1 ,include:

[0051] Step 1: Get the video frame image to be detected.

[0052] At this initial stage, the first task is to accurately extract the video frame images to be detected from the video stream. This process requires a high degree of timeliness and accuracy to ensure that the subsequent personnel identification and tracking tasks can be performed based on the latest and clear image data. Through advanced video decoding technology and frame synchronization mechanism, we can ensure that each frame of the captured image meets the analysis requirements, laying a solid foundation for the subsequent steps.

[0053] Step 2: Use the person recognition model to identify the video frame image to be detected and obtain the region of interest of the tracked person.

[0054] For image recognition, a person recognition model that has undergone deep learning and optimization training is used to perform detailed analysis and recognition on the extracted video frame images. The model can intelligently identify and accurately outline the area of ​​interest of the tracked person in the image. This process not only relies on the model's powerful feature extraction capabilities, but also combines contextual information and dynamic change analysis to improve the accuracy and robustness of recognition.

[0055] Step 3: Determine whether there is occlusion in the region of interest. If not, obtain the tracking result and update the tracking trajectory. If there is occlusion, perform an authenticity assessment on the region of interest based on the appearance features of the region of interest and the tracked person, and select different processing methods based on the authenticity assessment results.

[0056] After identifying the region of interest of the tracked person, it is necessary to further determine whether there is occlusion in the region. When occlusion is detected, the authenticity of the occluded region is evaluated based on the appearance characteristics of the region of interest and the tracked person to distinguish whether it is true occlusion or misjudgment and the probability of determining it as the tracked person in the case of occlusion. Based on the evaluation results, we will flexibly select processing strategies to minimize the impact of occlusion on tracking performance. Among them, the appearance characteristics of the tracked person include: clothing characteristics (clothing color, clothing style, accessories, etc.), body characteristics (height, weight, body shape, etc.), facial characteristics (facial contour, facial features, hairstyle and hair color, etc.), skin characteristics (skin color, skin color changes, etc.), behavioral characteristics (walking habits, posture changes, etc.), and other characteristics (carrying items, tattoos or scars, etc.).

[0057] Step 4: If the authenticity evaluation result is greater than the preset condition, the trajectory is predicted based on the tracking trajectory and the position of the tracked person, the predicted position and predicted time of the tracked person after the occlusion is eliminated are obtained, and the tracked person is matched at the predicted time and predicted position. If the tracked person is matched, the tracking trajectory is updated based on the predicted trajectory; if the tracked person is not matched, the current position of the tracked person is obtained, and the tracking trajectory is restored based on the current position of the tracked person and the occlusion position.

[0058] Step 5: If the authenticity evaluation result is less than or equal to the preset condition, perform occlusion recovery on the region of interest, and evaluate the authenticity of the region of interest after occlusion recovery based on the occlusion recovery result. If the authenticity evaluation result of the region of interest after occlusion recovery is greater than the preset condition, return to step 4; if the authenticity evaluation result of the region of interest after occlusion recovery is less than or equal to the preset condition, obtain the current position of the tracked person, and restore the tracking trajectory based on the current position of the tracked person and the occlusion position.

[0059] In another embodiment, determining whether the region of interest is blocked specifically includes:

[0060] According to the appearance features of the tracked person, multiple appearance feature key point areas are divided; the divided appearance feature key point areas should be representative and can retain sufficient recognition information even when blocked. For example, in the face area, key point areas such as eyes, nose, and mouth can be defined; in the body part, key points such as shoulders, waist, and knees can be selected; in the clothing area, clothing patterns can be selected for division.

[0061] Match the appearance feature key point area within the region of interest; use a matching algorithm (such as feature point matching, template matching, etc.) to match the previously defined appearance feature key point area, and quickly and accurately identify whether the key point area exists in the region of interest of the current frame.

[0062] The number of appearance feature key point regions matched in the region of interest is counted. If the number is less than the number of appearance feature key point regions, the region of interest is determined to be blocked.

[0063] In another embodiment, the authenticity of the region of interest is evaluated based on the appearance features of the region of interest and the tracked person, including:

[0064] Calculate the occlusion judgment value of the appearance feature key point area respectively; for each defined appearance feature key point area (such as eyes, nose, mouth, shoulders, etc.), we use the matching algorithm mentioned above to calculate its occlusion judgment value in the current frame. This value can be a percentage, indicating the degree of occlusion of the appearance feature key point area.

[0065] The importance index of the appearance feature key point area is determined based on the appearance features; different appearance feature key point areas have different importance for the recognition of the overall appearance. For example, in face recognition, the eyes and mouth are often more important than ears or hair because they are the main basis for expression and identity recognition. Therefore, it is necessary to assign an importance index to each appearance feature key point area based on the characteristics of the appearance feature key point area and its role in the overall appearance. This index can be a numerical value determined by expert scoring, data learning or other methods to quantify the importance of the appearance feature key point area.

[0066] According to the importance index, the occlusion judgment value is weighted and weighted summed. The authenticity evaluation result of the region of interest is obtained based on the weighted summation result. The result of the weighted summation can be directly used as an evaluation indicator of the authenticity of the region of interest. The higher this index is, the higher the authenticity and integrity of the region of interest is maintained under occlusion; conversely, it means that the region may be severely occluded and its authenticity is questioned. Based on this evaluation result, the system can respond accordingly, such as adjusting the tracking strategy or starting the occlusion recovery mechanism.

[0067] Furthermore, the occlusion judgment values ​​of the key point areas of the appearance features are calculated respectively, including:

[0068] For the unmatched appearance feature key point region, first appearance information of the unmatched appearance feature key point region and second appearance information of the appearance feature key point region corresponding to the region of interest are respectively obtained; based on the first appearance information, the second appearance information and the occlusion determination function, an occlusion determination value corresponding to the appearance feature key point region is determined;

[0069] For the matched appearance feature key point area, the occlusion determination value is determined to be zero.

[0070] In another embodiment, trajectory prediction is performed based on the tracking trajectory and the position of the tracked person, and the predicted position and predicted time of the tracked person after the occlusion is removed are obtained, including:

[0071] Obtain historical trajectory point information of the occluder, and determine whether the occluder is a static target based on the historical trajectory point information;

[0072] If the occluder is a static target, the turning probability of the tracked person is predicted based on the position information of the occluder and the position of the tracked person, and the uniform speed model of the tracked person is determined based on the tracking trajectory. The predicted position and predicted time of the tracked person after the occlusion is removed are predicted based on the position of the tracked person, the uniform speed model of the tracked person, and the turning probability.

[0073] If the occlusion is a dynamic target, the turning probability of the tracked person and the turning probability of the occlusion are predicted respectively based on the position information of the occlusion and the position of the tracked person, and the uniform speed model of the tracked person is determined based on the tracking trajectory, and the uniform speed model of the occlusion is determined according to the historical trajectory point information. Based on the position of the tracked person, the uniform speed model of the tracked person, the position of the occlusion, the uniform speed model of the occlusion, the turning probability of the tracked person, and the turning probability of the occlusion, the predicted position and predicted time of the tracked person after the occlusion is eliminated are predicted.

[0074] In another embodiment, performing occlusion recovery on the region of interest includes:

[0075] Use the semantic segmentation model to perform semantic segmentation on the region of interest, remove occluders, and obtain the occluder area;

[0076] Reduce the dimension of the occluder area and convert the RGB color space of the occluder area into the YCbCr color space;

[0077] Determine the contour point set of the occluded area, repair each pixel in the contour point set, and update the contour point set, removing the repaired pixels until the contour point set is empty;

[0078] After completing the repair of all pixels, the repaired area is obtained;

[0079] Convert the YCbCr color space of the repaired area back to RGB color space.

[0080] On the other hand, the present invention also provides a system for implementing the above-mentioned person tracking method based on authenticity and occlusion recovery, such as Figure 2 As shown, including:

[0081] An acquisition module, used for acquiring the video frame image to be detected;

[0082] A region extraction model is used to identify the video frame image to be detected using a person recognition model to obtain the region of interest of the tracked person;

[0083] An occlusion judgment module determines whether there is occlusion in the area of ​​interest;

[0084] An evaluation module is used to evaluate the authenticity of the region of interest based on the appearance characteristics of the region of interest and the tracked person if there is occlusion, and to determine different processing methods based on the authenticity evaluation results;

[0085] An update module, used to obtain the tracking result and update the tracking trajectory if there is no occlusion;

[0086] The personnel matching module is used to predict the track based on the tracking track and the position of the tracked person if the authenticity evaluation result is greater than the preset condition, obtain the predicted position and predicted time of the tracked person after the occlusion is eliminated, and match the tracked person at the predicted time and predicted position. If the tracked person is matched, the tracking track is updated based on the predicted track.

[0087] The occlusion recovery module is used to perform occlusion recovery on the region of interest if the authenticity evaluation result is less than or equal to the preset condition, and evaluate the authenticity of the region of interest after occlusion recovery based on the occlusion recovery result. If the region of interest after occlusion recovery is greater than the preset value, it returns to the person matching module;

[0088] The recovery tracking module obtains the current position of the tracked person and recovers the tracking trajectory based on the current position of the tracked person and the occlusion position if the tracked person is not matched or the area of ​​interest after occlusion recovery is less than or equal to the preset condition.

[0089] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0090] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A person tracking method based on authenticity and occlusion recovery, characterized in that: include: Step 1: Obtain the video frame image to be detected; Step 2: Using a personnel recognition model to recognize the video frame image to be detected, and obtaining a region of interest of the tracked person; Step 3: Determine whether the region of interest is blocked, and if not, obtain the tracking result and update the tracking trajectory; If there is occlusion, the authenticity of the region of interest is evaluated according to the appearance features of the region of interest and the tracked person, and different processing methods are selected according to the authenticity evaluation results; Step 4: If the authenticity evaluation result is greater than the preset condition, the trajectory is predicted based on the tracking trajectory and the position of the tracked person, the predicted position and predicted time of the tracked person after the occlusion is eliminated are obtained, and the tracked person is matched at the predicted time and the predicted position. If the tracked person is matched, the tracking trajectory is updated based on the predicted trajectory; if the tracked person is not matched, the current position of the tracked person is obtained, and the tracking trajectory is restored based on the current position of the tracked person and the occlusion position; Step 5: If the authenticity evaluation result is less than or equal to the preset condition, perform occlusion recovery on the region of interest, and evaluate the authenticity of the region of interest after occlusion recovery based on the occlusion recovery result; if the authenticity evaluation result of the region of interest after occlusion recovery is greater than the preset condition, return to step 4; If the authenticity evaluation result of the region of interest after the occlusion restoration is less than or equal to a preset condition, the current position of the tracked person is obtained, and the tracking trajectory is restored based on the current position of the tracked person and the occlusion position.

2. The method for tracking people based on authenticity and occlusion recovery according to claim 1, characterized in that: Determining whether the region of interest is blocked specifically includes: Divide multiple appearance feature key point areas according to the appearance features of the tracked person; Matching the appearance feature key point region within the region of interest; The number of the appearance feature key point regions matched in the region of interest is counted, and if the number is less than the number of the appearance feature key point regions, it is determined that the region of interest is blocked.

3. The method for tracking people based on authenticity and occlusion recovery according to claim 2, characterized in that: Performing authenticity evaluation on the region of interest according to appearance features of the region of interest and the tracked person, including: Calculating the occlusion determination values ​​of the appearance feature key point areas respectively; Determining the importance index of the appearance feature key point area based on the appearance feature; The occlusion determination value is weighted according to the importance index, and a weighted sum is performed, and the authenticity evaluation result of the region of interest is obtained according to the weighted sum result.

4. The method for tracking people based on authenticity and occlusion recovery according to claim 3, characterized in that: Calculating the occlusion determination values ​​of the appearance feature key point areas respectively, specifically including: For an unmatched appearance feature key point region, first appearance information of the unmatched appearance feature key point region and second appearance information of the appearance feature key point region corresponding to the region of interest are respectively obtained; based on the first appearance information, the second appearance information and an occlusion determination function, an occlusion determination value corresponding to the appearance feature key point region is determined; For the matched appearance feature key point area, the occlusion determination value is determined to be zero.

5. The method for tracking people based on authenticity and occlusion recovery according to claim 1, characterized in that: The track prediction is performed based on the tracked track and the position of the tracked person, and the predicted position and predicted time of the tracked person after the occlusion is eliminated are obtained, including: Obtain historical trajectory point information of the occluder, and determine whether the occluder is a static target based on the historical trajectory point information; If the occluder is a static target, the turning probability of the tracked person is predicted based on the position information of the occluder and the position of the tracked person, and the uniform speed model of the tracked person is determined based on the tracking trajectory, and the predicted position and predicted time of the tracked person after the occlusion is eliminated are predicted based on the position of the tracked person, the uniform speed model of the tracked person, and the turning probability; If the obstruction is a dynamic target, the turning probability of the tracked person and the turning probability of the obstruction are predicted respectively based on the position information of the obstruction and the position of the tracked person, and the uniform speed model of the tracked person is determined based on the tracking trajectory, and the uniform speed model of the obstruction is determined according to the historical trajectory point information. Based on the position of the tracked person, the uniform speed model of the tracked person, the position of the obstruction, the uniform speed model of the obstruction, the turning probability of the tracked person, and the turning probability of the obstruction, the predicted position and predicted time of the tracked person after the occlusion is eliminated are predicted.

6. The method for tracking people based on authenticity and occlusion recovery according to claim 1, characterized in that: Performing occlusion recovery on the region of interest, including: Performing semantic segmentation on the region of interest using a semantic segmentation model, removing occluders, and obtaining an occluder region; Reducing the dimension of the occluder region, and converting the RGB color space of the occluder region into a YCbCr color space; Determine a contour point set of the occluder area, repair each pixel in the contour point set, and update the contour point set, removing the repaired pixel points until the contour point set is empty; After completing the repair of all pixels, the repaired area is obtained; Convert the YCbCr color space of the repaired area back to RGB color space.

7. A system for implementing a person tracking method based on authenticity and occlusion recovery as claimed in any one of claims 1 to 6, characterized in that: include: An acquisition module, used for acquiring the video frame image to be detected; A region extraction model is used to identify the video frame image to be detected using a person recognition model to obtain a region of interest for tracking the person; An occlusion judgment module is used to judge whether there is occlusion in the region of interest; An evaluation module, for evaluating the authenticity of the region of interest according to the appearance features of the region of interest and the tracked person if there is occlusion, and determining to select different processing methods according to the authenticity evaluation result; An update module, used to obtain the tracking result and update the tracking trajectory if there is no occlusion; A personnel matching module is used to predict the track based on the tracking track and the position of the tracked person if the authenticity evaluation result is greater than a preset condition, obtain the predicted position and predicted time of the tracked person after the occlusion is eliminated, and match the tracked person at the predicted time and the predicted position. If the tracked person is matched, the tracking track is updated based on the predicted track; If the tracked person is not matched, the current position of the tracked person is obtained, and the tracking trajectory is restored based on the current position of the tracked person and the occlusion position; An occlusion recovery module is used to perform occlusion recovery on the region of interest if the authenticity evaluation result is less than or equal to a preset condition, evaluate the authenticity of the region of interest after occlusion recovery based on the occlusion recovery result, and return to the person matching module if the region of interest after occlusion recovery is greater than a preset value; The recovery tracking module obtains the current position of the tracked person if the tracked person is not matched or the area of ​​interest after the occlusion is restored is less than or equal to a preset condition, and recovers the tracking trajectory based on the current position of the tracked person and the occlusion position.

Images