Accurate and rapid light following method

Abstract

The invention discloses an accurate and rapid light following method, which comprises the following steps: firstly, finding a target person from a field picture through face recognition, then judging the confidence coefficient of the real-time position of the target person, when the confidence coefficient is lower than a first preset value, marking the next frame to perform face recognition to track the target person, and if the face recognition fails or is not marked, judging that the target person is not marked. If yes, the predicted position of the target person is directly matched with the real-time positions of all persons in the next frame, the matched corresponding persons serve as the target person, then target person tracking of the next frame is carried out, and when the corresponding persons are not matched, the predicted position of the target person serves as the real-time position of the target person; and then face recognition is carried out in the next frame, and the real-time position of the target person is tracked by light. According to the method, position tracking is taken as a main part, and whether face recognition is carried out in a next frame of on-site picture is determined in combination with judgment on the real-time position confidence of the current frame, so that both the rapidity of position tracking and the accuracy of face recognition are taken into account.

Description

Technical Field

[0001] The present invention relates to the technical field of stage lights, and more specifically, to a precise and fast light chasing method. Background Art

[0002] In the current stage lighting system, there are generally three ways for the light to follow the person on the stage: one is that the lighting technician manually controls the movement of the light to track a specific person for lighting. Since the lamps are generally far away from the stage, the manual tracking accuracy is greatly affected by human factors and is prone to jitter; another common method is for stage actors to wear special tags, such as UWB positioning tags, infrared marker tags, etc., but on the one hand, it may be inconvenient for actors to wear tags in certain occasions, and on the other hand, the data transmission of the tag may be interfered by the human body or the surrounding scene, resulting in inaccurate positioning or positioning jitter; another method is to use machine vision to identify the target person and track it, but the machine vision recognition speed is slow, and the target person is easy to lose the target when it is blocked. Therefore, an accurate and fast light chasing method is needed. Summary of the invention

[0003] In order to overcome at least one of the defects of the prior art described above, the present invention provides an accurate and fast light chasing method, which uses a purely visual method to quickly and accurately lock a target person.

[0004] In order to solve the above technical problems, the technical solution adopted by the present invention is: a precise and fast light chasing method, comprising the following steps: S1, continuously detecting the scene captured by the camera until a target person with specific facial features is identified, and obtaining the real-time position and movement information of the target person; S2, the light of the lamp tracks the real-time position of the target person, and determines whether the confidence of the real-time position is less than a first preset value, and if so, marks that face recognition needs to be performed in the next frame; S3, obtaining the predicted position of the target person in the next frame according to the real-time position and motion information of the target person; S4, the camera captures the next frame of the scene, and obtains the real-time position and face position of all characters in the next frame of the scene; S5, when face recognition is required in the next frame, the faces of all the characters in the next frame of the scene in step S4 are recognized. If the target person is recognized, the real-time position of the target person in the scene is updated, and it is marked that face recognition is not required in the next frame, and then jump to step S2. If the target person is not recognized, jump to step S6; when it is marked in step S2 that face recognition is not required in the next frame, jump directly to step S6; S6. Match the predicted position of the target person in the next frame in step S3 with the real-time positions of all characters in the next frame of the scene in step S4. If the match is successful, the matched corresponding person is considered to be the target person, the real-time position of the target person in this scene is updated, and the process jumps to step S2. If the match is unsuccessful, the predicted position of the target person in step S3 is considered to be the real-time position of the target person in the next frame of the scene, and the motion information remains unchanged, and the confidence of the real-time position is considered to be less than the first preset value, and then the process jumps to step S2.

[0005] The precise and fast light tracking method first finds the target person from the scene picture through face recognition, and then determines the confidence of the real-time position of the target person. When the confidence is lower than a first preset value, it is marked that face recognition is required to assist in tracking the target person in the next frame of the scene. If face recognition fails or is not marked, the predicted position of the target person in the next frame is directly matched with the real-time positions of all characters in the next frame of the scene. When the corresponding person is matched, it is used as the target person, and the real-time position of the target person is updated and its confidence is determined, and then the target person tracking in the next frame is entered. When the corresponding person is not matched, the predicted position of the target person in the next frame is used as the real-time position of the target person in the next frame of the scene, and its motion information is considered unchanged. Then face recognition is performed in the next frame to track the target person, and the light of the lamp tracks the real-time position of the target person.

[0006] This method is mainly based on position tracking, and predicts the position of the target person in the next frame. It also combines the judgment of the real-time position confidence of the current frame to decide whether to perform face recognition in the next frame of the scene to assist in tracking the target person. It takes into account the rapidity of position tracking and the accuracy of face recognition, thereby achieving accurate and fast light chasing.

[0007] Furthermore, in step S6, the number of consecutive matching failures when matching the predicted position of the target person in step S3 in the next frame with the real-time positions of all characters in the next frame of the scene in step S4 is counted. When it exceeds the second preset value, the target person is considered lost and the process returns to step S1. In this way, in step S6, the target person is not considered lost immediately after the matching fails, but is considered lost only after multiple confirmations, and complex calculations are performed to find the target person again through face recognition, which saves computing power, speeds up the light chasing speed, and avoids the slow light chasing speed caused by the face recognition being performed immediately without recognizing the target person due to occasional occlusion or poor light.

[0008] Furthermore, an initialization step is also included before step S1. In the initialization step, the specific facial features are obtained by inputting a photo or manually selecting a target person from the scene. When the specific facial features are obtained by inputting a photo, the process jumps to step S1. When the specific facial features are obtained by manually selecting a target person from the scene, it is considered that the target person is directly identified in step S1, and the real-time position and motion information of the target person is obtained, and then the process jumps to step S2. To obtain specific facial features by inputting a photo, it is necessary to first compare the specific facial features with the facial information of all the people in the scene before the target person can be found. To obtain specific facial features by manually selecting the target person from the scene, it is equivalent to directly giving the real-time position and motion information of the target person. The comparison process of the specific facial features with the facial information of all the people in the scene is completed by the human brain.

[0009] Furthermore, the order of step S3 and step S4 can be swapped at will. The contents of step S3 and step S4 can be completed independently of each other without interfering with each other, so the order can be swapped, but the jumps between step S3 and step S4 in the entire method also need to be modified accordingly.

[0010] Furthermore, in step S3, when the predicted position of the target person in the next frame is obtained based on the real-time position and motion information of the target person, Kalman filtering is used. Kalman filter is a recursive state estimation algorithm that provides the best state estimation result by continuously estimating and updating the state of the system, that is, it continuously iterates the process of "making predictions-updating the predicted values ​​to the optimal values ​​based on the measured values". The technology is mature and suitable for tracking stable moving targets.

[0011] Furthermore, the real-time position is determined based on the recognized character frame, and the confidence of the real-time position is the confidence of the character frame. Any point in the character frame can be selected as the real-time position; judging the confidence of the real-time position is to judge the confidence of the character frame, so as to confirm whether it is blocked and how reliable the detection result is.

[0012] Furthermore, the character frame refers to a head frame, which is smaller, has fewer postures, is more accurately detected, and is less prone to shaking.

[0013] Furthermore, when detecting people in step S1 and step S4, the backbone network is used to extract features from the scene and form a multi-scale feature pyramid. Each scale feature layer outputs 5 branches: character frame confidence, character frame position, face confidence, face position, and at least 3 face key points. Each feature layer corresponds to several anchor points to adapt to target people of different proportions. Then, the character frame confidence is filtered, and all outputs greater than the third preset value are retained. After that, the character frame position is combined with the anchor point information to calculate the real-time position of the corresponding person in the scene, and non-maximum suppression is performed to obtain the final real-time position of all characters. Then, the face confidence corresponding to all detected people is filtered, and all outputs greater than the fourth preset value are retained. Finally, the face position and at least 3 face key points are combined with the anchor point information to calculate the face position in the scene. This solution detects the character frame and the face at the same time, uses the character frame detection result as the real-time position, and uses the face feature as the ReID information. It completes the detection of the character frame and the face at the same time, which greatly improves the detection speed.

[0014] Furthermore, the camera is mounted on the lamp head of the lamp for emitting a light beam, and rotates with the lamp head, so that the camera can rotate to find the target person, and the synchronous rotation can facilitate the light of the lamp to illuminate the target person after the camera finds the target person.

[0015] Furthermore, when the light of the lamp tracks the real-time position of the target person, the head of the target person is placed at the center point of the camera live picture. This ensures that no matter which direction the target person moves, he will not disappear from the live picture in a short time, which is convenient for tracking.

[0016] Furthermore, in step S1, the camera will follow the rotation of the lamp head to find the target person while continuously capturing the live scene, so that the target person can be found by the camera regardless of where he is on the stage, which is more intelligent.

[0017] Furthermore, in step S6, the predicted position of the target person in the next frame in step S3 is matched with the real-time position of all characters in the next frame of the scene in step S4, and the Hungarian algorithm is used for matching. The IoU between the predicted position of the target person and the real-time position of all characters is calculated, and the Hungarian algorithm is used for matching in combination with the confidence of the real-time position and the IoU. The technology is mature and the matching is accurate.

[0018] Furthermore, in step S1 and step S5, when identifying a target person with specific facial features, it is necessary to compare the specific facial features with the facial information of all persons, and select the corresponding person with the greatest similarity and greater than the fifth preset value as the target person, otherwise it is considered that the target person is not found, so as to ensure that the probability of the selected corresponding person being the target person is the highest and basically meets the requirements. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a flow chart of the accurate and fast light chasing method of the present invention.

[0020] Figure 2 It is a schematic diagram of the structure of a system using the precise and fast light chasing method of the present invention.

[0021] Figure 3 It is a schematic diagram of the process of detecting a person according to the present invention.

[0022] Figure 4 It is a schematic diagram of the process of face recognition of a target person in the present invention. DETAILED DESCRIPTION

[0023] The drawings are only for illustrative purposes and cannot be construed as limiting the present invention. To better illustrate the present embodiment, some parts of the drawings may be omitted, enlarged, or reduced, and do not represent the size of the actual product. For those skilled in the art, it is understandable that some well-known structures and their descriptions may be omitted in the drawings. The positional relationships described in the drawings are only for illustrative purposes and cannot be construed as limiting the present invention.

[0024] like Figure 1 to Figure 2 The present invention provides a precise and fast light chasing method, comprising the following steps: S1, continuously detecting the scene captured by the camera until a target person with specific facial features is identified, and obtaining the real-time position and movement information of the target person; S2, the light of the lamp tracks the real-time position of the target person, and determines whether the confidence of the real-time position is less than a first preset value, and if so, marks that face recognition needs to be performed in the next frame; S3, obtaining the predicted position of the target person in the next frame according to the real-time position and motion information of the target person; S4, the camera captures the next frame of the scene, and obtains the real-time position and face position of all characters in the next frame of the scene; S5, when face recognition is required in the next frame, the faces of all the characters in the next frame of the scene in step S4 are recognized. If the target person is recognized, the real-time position of the target person in the scene is updated, and it is marked that face recognition is not required in the next frame, and then jump to step S2. If the target person is not recognized, jump to step S6; when it is marked in step S2 that face recognition is not required in the next frame, jump directly to step S6; S6. Match the predicted position of the target person in the next frame in step S3 with the real-time positions of all characters in the next frame of the scene in step S4. If the match is successful, the matched corresponding person is considered to be the target person, the real-time position of the target person in this scene is updated, and the process jumps to step S2. If the match is unsuccessful, the predicted position of the target person in step S3 is considered to be the real-time position of the target person in the next frame of the scene, and the motion information remains unchanged, and the confidence of the real-time position is considered to be less than the first preset value, and then the process jumps to step S2.

[0025] The precise and fast light tracking method first finds the target person from the scene picture through face recognition, and then determines the confidence of the real-time position of the target person. When the confidence is lower than a first preset value, it is marked that face recognition is required to assist in tracking the target person in the next frame of the scene. If face recognition fails or is not marked, the predicted position of the target person in the next frame is directly matched with the real-time positions of all characters in the next frame of the scene. When the corresponding person is matched, it is used as the target person, and the real-time position of the target person is updated and its confidence is determined, and then the target person tracking in the next frame is entered. When the corresponding person is not matched, the predicted position of the target person in the next frame is used as the real-time position of the target person in the next frame of the scene, and its motion information is considered unchanged. Then face recognition is performed in the next frame to track the target person, and the light of the lamp tracks the real-time position of the target person.

[0026] This method is mainly based on position tracking, and predicts the position of the target person in the next frame. It also combines the judgment of the real-time position confidence of the current frame to decide whether to perform face recognition in the next frame of the scene to assist in tracking the target person. It takes into account the rapidity of position tracking and the accuracy of face recognition, thereby achieving accurate and fast light chasing.

[0027] It should be noted that as long as it is marked in step S2 that face recognition needs to be performed in the next frame, no matter whether the confidence of the real-time position of the target person is greater than or equal to the first preset value in the subsequent steps, face recognition needs to be performed in the next frame after jumping to step S2, unless the mark that face recognition needs to be performed in the next frame is eliminated in the subsequent steps.

[0028] In a preferred embodiment of the present invention, in step S6, the predicted position of the target person in the next frame in step S3 is matched with the real-time position of all characters in the next frame of the scene in step S4. When the number of consecutive matching failures exceeds the second preset value, the target person is considered lost and the process returns to step S1. In this way, in step S6, the target person is not considered lost immediately after the matching fails, but is considered lost only after multiple confirmations, and complex calculations are performed to find the target person again through face recognition, which saves computing power, speeds up the light chasing speed, and avoids the slow light chasing speed caused by the face recognition being performed immediately without identifying the target person due to occasional occlusion or poor light.

[0029] The second preset value may be 2, 3, 4, 5, 6, 7, 8, 9 or 10, preferably 5.

[0030] In a preferred embodiment of the present invention, an initialization step is also included before step S1. In the initialization step, the specific facial features are obtained by inputting a photo or manually selecting a target person from the scene. When the specific facial features are obtained by inputting a photo, the process jumps to step S1. When the specific facial features are obtained by manually selecting a target person from the scene, it is considered that the target person is directly identified in step S1, and the real-time position and motion information of the target person is obtained, and then the process jumps to step S2. To obtain specific facial features by inputting a photo, it is necessary to first compare the specific facial features with the facial information of all the people in the scene before the target person can be found. To obtain specific facial features by manually selecting the target person from the scene, it is equivalent to directly providing the real-time position and motion information of the target person. The comparison process of the specific facial features with the facial information of all the people in the scene is completed by the human brain.

[0031] In the present application, the specific facial features are preferably obtained by inputting photos, and the photos are sent to the lamps through the control server.

[0032] In a preferred embodiment of the present invention, the order of step S3 and step S4 can be swapped at will. The contents of step S3 and step S4 can be completed independently of each other without interfering with each other, so the order can be swapped, but the jumps of step S3 and step S4 in the entire method also need to be modified accordingly.

[0033] In a preferred embodiment of the present invention, in step S3, when obtaining the predicted position of the target person in the next frame based on the real-time position and motion information of the target person, Kalman filtering is used. Kalman filter is a recursive state estimation algorithm, which provides the best state estimation result by continuously estimating and updating the state of the system, that is, continuously iterating the process of "making predictions-updating the predicted values ​​to the optimal values ​​according to the measured values". The technology is mature and suitable for tracking stable moving targets. In the present application, obtaining the predicted position of the target person in the next frame based on the real-time position and motion information of the target person is "making predictions", updating the real-time position of the target person according to the real-time position of the target person identified by face recognition in the next frame or the real-time position of the corresponding person obtained by matching the predicted position with the real-time positions of all characters, or forcibly considering that the predicted position of the target person in step S3 is the real-time position of the target person in the next frame of the scene, that is, "updating the predicted value to the optimal value according to the measured value".

[0034] In a preferred embodiment of the present invention, the real-time position is determined based on the recognized character frame, and the confidence of the real-time position is the confidence of the character frame. Any point in the character frame can be selected as the real-time position; judging the confidence of the real-time position is to judge the confidence of the character frame, so as to confirm whether it is blocked and how reliable the detection result is.

[0035] The character frame may be a body frame or a head frame. In a preferred embodiment of the present invention, the character frame refers to a head frame. Compared with the body frame, the head frame is smaller, has fewer postures, is more accurately detected, and is less prone to shaking.

[0036] like Figure 3In a preferred embodiment of the present invention, when detecting a person in step S1 and step S4, the backbone network is used to extract features from the scene and form a multi-scale feature pyramid. Each scale feature layer outputs 5 branches: character frame confidence, character frame position, face confidence, face position, and at least 3 face key points. Each feature layer corresponds to several anchor points to adapt to target people of different proportions. Then, the character frame confidence is filtered, and all outputs greater than the third preset value are retained. Then, the character frame position is combined with the anchor point information to calculate the real-time position of the corresponding person in the scene, and non-maximum suppression is performed to obtain the final real-time position of all characters. Then, the face confidence corresponding to all detected people is filtered, and all outputs greater than the fourth preset value are retained. Finally, the face position and at least 3 face key points are combined with the anchor point information to calculate the face position in the scene. Compared with the prior art, it is necessary to first perform character detection on the scene, and then perform face detection on all character frames. At this time, if there are many character frames for face detection, it will be very time-consuming. This solution detects person frames and faces at the same time, uses the person frame detection results as the real-time position, and uses facial features as ReID information. It completes the detection of person frames and faces at the same time, greatly improving the detection speed.

[0037] The five branches, person frame confidence, person frame position, face confidence, face position, and at least three facial key points, respectively predict whether a person is detected by the feature point, the person position, whether a face is detected, the face position, and the positions of at least three facial key points.

[0038] Preferably, the number of facial key points is 5.

[0039] In this application, the backbone network is a ResNet50 network, and then FPN is used to form a multi-scale feature pyramid.

[0040] After obtaining the real-time positions of all people in the scene, filter (some people may have their backs to the camera and their faces cannot be observed) to get those people whose faces can be seen by the camera, and obtain their face positions. When performing face recognition, only the corresponding area needs to be recognized.

[0041] The aforementioned person detection model also needs to be trained. The training data contains label information of person, face, and at least 3 key points. The model training adopts a training strategy similar to RetinaNet. The FocalLoss loss function is used for person frame confidence and face confidence, and the smooth-L1 loss function is used for person frame position, face position, and at least 3 key points. It should be noted that when the person frame is a head frame, the ground turth distribution of the person frame position and face position are both distributed using the center of the head frame.

[0042] In a preferred embodiment of the present invention, the camera is mounted on the lamp head of the lamp for emitting a light beam, and rotates with the lamp head, so that the camera can rotate to find the target person, and the synchronous rotation can facilitate the light of the lamp to illuminate the target person after the camera finds the target person.

[0043] It should be noted that if the camera is installed on the lamp head of the lamp for emitting a light beam and rotates with the lamp head, then when calculating the predicted position of the target person, the displacement of the camera following the movement of the lamp head between two frames needs to be considered.

[0044] In a preferred embodiment of the present invention, when the light of the lamp tracks the real-time position of the target person, the head of the target person is placed at the center point of the camera live picture. This ensures that no matter which direction the target person moves, he will not disappear from the live picture in a short time, which is convenient for tracking.

[0045] Of course, the head of the target person may be appropriately offset relative to the center point of the camera scene image to ensure that the light beam from the lamp head is irradiated to the face or feet of the target person.

[0046] In a preferred embodiment of the present invention, in step S1, the camera will follow the rotation of the lamp head to find the target person while continuously capturing the scene, so that the target person can be found by the camera regardless of where he is on the stage, which is more intelligent.

[0047] In a preferred embodiment of the present invention, in step S6, when matching the predicted position of the target person in the next frame in step S3 with the real-time positions of all characters in the next frame of the scene in step S4, the Hungarian algorithm is used for matching. The IoU between the predicted position of the target person and the real-time positions of all characters is calculated, and the Hungarian algorithm is used for matching in combination with the confidence of the real-time position and the IoU. The technology is mature and the matching is accurate.

[0048] like Figure 4 In a preferred embodiment of the present invention, in step S1 and step S5, when identifying a target person with specific facial features, it is necessary to compare the specific facial features with the facial information of all persons, and select the corresponding person with the greatest similarity and greater than the fifth preset value as the target person, otherwise it is considered that the target person is not found. To ensure that the probability that the selected corresponding person is the target person is the highest and basically meets the requirements, if the similarity is less than or equal to the fifth preset value, it is considered that the target person is not identified.

[0049] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the embodiments here. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the claims of the present invention.

Claims

1. A precise and fast light chasing method, characterized in that: The following steps are involved: S1, continuously detecting the scene captured by the camera until a target person with specific facial features is identified, and obtaining the real-time position and movement information of the target person; S2, the light of the lamp tracks the real-time position of the target person, and determines whether the confidence of the real-time position is less than a first preset value, and if so, marks that face recognition needs to be performed in the next frame; S3, obtaining the predicted position of the target person in the next frame according to the real-time position and motion information of the target person; S4, the camera captures the next frame of the scene, and obtains the real-time position and face position of all characters in the next frame of the scene; S5, when face recognition is required in the next frame, the faces of all the characters in the next frame of the scene in step S4 are recognized. If the target person is recognized, the real-time position of the target person in the scene is updated, and it is marked that face recognition is not required in the next frame, and then jump to step S2. If the target person is not recognized, jump to step S6; When it is marked in step S2 that face recognition is not required in the next frame, jump directly to step S6; S6. Match the predicted position of the target person in the next frame in step S3 with the real-time positions of all characters in the next frame of the scene in step S4. If the match is successful, the matched corresponding person is considered to be the target person, the real-time position of the target person in this scene is updated, and the process jumps to step S2. If the match is unsuccessful, the predicted position of the target person in step S3 is considered to be the real-time position of the target person in the next frame of the scene, and the motion information remains unchanged, and the confidence of the real-time position is considered to be less than the first preset value, and then the process jumps to step S2.

2. The accurate and fast light chasing method according to claim 1, characterized in that: In step S6, the number of consecutive matching failures when matching the predicted position of the target person in step S3 in the next frame with the real-time positions of all characters in the next frame of the scene in step S4 is counted. When the number exceeds a second preset value, it is considered that the target person is lost and the process returns to step S1.

3. The accurate and fast light chasing method according to claim 1, characterized in that: Before step S1, an initialization step is also included. In the initialization step, specific facial features are obtained by inputting photos or manually selecting the target person from the scene. When the specific facial features are obtained by inputting photos, the process jumps to step S1. When the specific facial features are obtained by manually selecting the target person from the scene, it is considered that the target person is directly identified in step S1, and the real-time position and motion information of the target person is obtained, and then the process jumps to step S2.

4. The accurate and fast light chasing method according to claim 1, characterized in that: The order of step S3 and step S4 can be changed at will.

5. The accurate and fast light chasing method according to claim 1, characterized in that: In step S3, Kalman filtering is used to obtain the predicted position of the target person in the next frame based on the real-time position and motion information of the target person.

6. The accurate and fast light chasing method according to claim 1, characterized in that: The real-time position is determined based on the recognized character frame, and the confidence of the real-time position is the confidence of the character frame.

7. The accurate and fast light chasing method according to claim 6, characterized in that: The character frame refers to a head frame.

8. The precise and fast light chasing method according to claim 6 or 7, characterized in that: When detecting a person in step S1 and step S4, the backbone network is used to extract features of the scene and form a multi-scale feature pyramid. The feature layer of each scale outputs 5 branches: person frame confidence, person frame position, face confidence, face position, and at least 3 face key points. Each feature layer corresponds to several anchor points to adapt to target persons of different proportions. The person frame confidence is then filtered, and all outputs greater than a third preset value are retained. The person frame position is then combined with the anchor point information to calculate the real-time position of the corresponding person in the scene, and non-maximum suppression is performed to obtain the final real-time position of all the characters. The face confidences corresponding to all detected persons are then filtered, and all outputs greater than a fourth preset value are retained. Finally, the face position and at least 3 face key points are combined with the anchor point information to calculate the face position in the scene.

9. The accurate and fast light chasing method according to claim 1, characterized in that: The camera is mounted on the lamp head of the lamp for emitting a light beam and rotates with the lamp head.

10. The accurate and fast light chasing method according to claim 9, characterized in that: When the light of the lamp tracks the real-time position of the target person, the head of the target person is placed at the center point of the camera's live picture.

11. The accurate and fast light chasing method according to claim 9, characterized in that: In step S1, while the camera continues to capture the scene, it will follow the rotation of the light head to find the target person.

12. The accurate and fast light chasing method according to claim 1, characterized in that: In step S6, the predicted position of the target person in the next frame in step S3 is matched with the real-time positions of all the characters in the next frame of the scene in step S4, and the Hungarian algorithm is used for matching.

13. The accurate and fast light chasing method according to claim 1, characterized in that: In steps S1 and S5, when identifying a target person with specific facial features, it is necessary to compare the specific facial features with the facial information of all persons, and select the corresponding person with the greatest similarity and greater than the fifth preset value as the target person, otherwise it is considered that the target person is not found.

Images