A real-time multi-target tracking method for passengers in terminal

Abstract

The present invention relates to the technical field of video analysis means, and specifically, to a method for real-time multi-target tracking of passengers in a terminal. This method is based on video surveillance data in the terminal, with passengers as the main tracking objects, and aims to complete real-time multi-target tracking in the terminal scenario. Secondly, the proposed multi-target tracking method can analyze the surveillance video in the terminal in real time and accurately track multiple passengers. This technology not only improves monitoring efficiency, such as being able to count passengers in real time and set early warning broadcasts, but also optimizes passenger flow lines, reduces congestion, and thus enhances the overall travel experience of passengers. Combined with the current aviation transportation industry's growing demand for safety and efficiency, this method will provide airport management departments with strong data and technical support, enabling them to respond to emergencies in a timely manner on the basis of real-time monitoring, and enhance the level of intelligence in airport operations.

Description

Technical Field

[0001] The present invention relates to the technical field of video analysis means, and in particular to a real-time multi-target tracking method for passengers in an airport terminal. Background Art

[0002] The shortcomings of existing multi-target tracking solutions applied in airport terminals are as follows:

[0003] Occlusion and target overlap: Within the terminal, passengers experience significant occlusion and interaction, such as luggage carts, seats, and even passenger overlap, which can lead to noise in target detection. Furthermore, due to the high level of passenger overlap, traditional multi-target tracking algorithms struggle to accurately distinguish adjacent passengers, making it difficult to associate subsequent targets with their trajectories. This can lead to ID swaps and track loss. This issue is particularly acute during crowded periods, severely impacting tracking accuracy.

[0004] Airport terminal scenes typically include multiple distinct areas, such as waiting areas, security checkpoints, and boarding gates. Existing multi-target tracking algorithms are often trained for specific scenarios, requiring high data acquisition and struggling to effectively handle complex background changes.

[0005] In airport terminal scenarios, passenger density is high, and the frame rate of surveillance video typically reaches 25-30 frames per second. However, existing multi-target tracking algorithms often rely on complex network structures to improve accuracy, which reduces the frame rate of processed video and makes it difficult to meet the needs of real-time monitoring. Summary of the Invention

[0006] The purpose of the present invention is to provide a real-time multi-target tracking method for passengers in an airport terminal to solve the above-mentioned problems in the prior art.

[0007] The present invention is achieved through the following technical solutions:

[0008] A real-time multi-target tracking method for passengers in an airport terminal, comprising:

[0009] The confidence level of the passenger detection set in the target video frame is obtained using the YOLOX detector, and a confidence threshold is set. The passenger detection set with a confidence level greater than the confidence threshold is divided into the first passenger detection set, and the passenger detection set with a confidence level less than the confidence threshold is divided into the second passenger detection set.

[0010] Obtain a set of historical passenger trajectories, obtain a set of passenger trajectory prediction values ​​based on a Kalman filter, establish an evaluation index based on the relationship between the speed and direction of the historical passenger trajectories and the height change of the trajectory detection box, establish an initial association cost based on the trajectory detection box and the evaluation index, and establish an association cost between the first passenger detection set and the passenger trajectory prediction value set based on the initial association cost to obtain a first cost matrix;

[0011] Applying the Hungarian algorithm to the first cost matrix yields a first set of successfully matched passenger trajectories, a second set of unmatched passenger trajectories, and a third set of unmatched passenger detections.

[0012] Calculate a second cost matrix for the second passenger detection set and the set of passenger trajectory prediction values ​​for the second passenger trajectory set. Apply the Hungarian algorithm to the second cost matrix to obtain a third set of successfully matched passenger trajectories and a fourth set of passenger trajectories that do not match any detections.

[0013] The target set is formed by obtaining the last matching detection in the fourth passenger trajectory set. The associated cost between the third passenger detection set and the target set is calculated to obtain a third cost matrix. The Hungarian algorithm is applied to the third cost matrix to output the fifth passenger trajectory set that successfully matches, the sixth passenger trajectory set that does not match any detection, and the fourth passenger detection set that does not match any trajectory.

[0014] Preferably, it also includes:

[0015] Obtain a combined total set of the first, third, and fifth passenger trajectory sets, and use the YOLOX detector to detect changes in the height of the historical trajectory detection frames to correct the trajectories in the total set;

[0016] The unreliability of the passenger detection value associated with the trajectory in the total set is calculated based on the detection values ​​of the previous k frames of the trajectory in the total set, and the passenger detection value is corrected based on the unreliability.

[0017] The predicted trajectory is updated using the corrected passenger detection value, and the final predicted trajectory is output.

[0018] Preferably, the step of obtaining a set of historical passenger trajectories and obtaining a set of passenger trajectory prediction values ​​based on a Kalman filter includes:

[0019]

[0020]

[0021] Where, for The state prediction value at the moment, is the predicted change matrix, is a track in the historical passenger track set, is the forecast error covariance matrix, is the updated forecast error covariance matrix, is the transposed matrix of the predicted change matrix, is the noise matrix.

[0022] Preferably, establishing an evaluation index based on the relationship between the speed direction of the passenger's historical trajectory and the height change of the trajectory detection frame includes:

[0023]

[0024] Where, As the evaluation index, Prediction box height, is the height of the detection box.

[0025] Preferably, the association cost for establishing the first passenger detection set and the passenger trajectory prediction value set based on the initial association cost includes:

[0026]

[0027] Where, is the association cost between the first passenger detection set and the passenger trajectory prediction value set, For the first passenger detection collection, is the set of passenger trajectory prediction values, is the cost of the evaluation indicator, is the first weight parameter, is the cost of the orientation consistency between the detection box and the detection box on the trajectory, is the second weight parameter, is the detection box confidence cost, is the third weight parameter, is the initial association cost.

[0028] Preferably, the passenger trajectory prediction value set of the second passenger detection set and the second cost matrix of the fourth passenger trajectory set include:

[0029]

[0030] Where, is the association cost between the second passenger detection set and the passenger trajectory prediction value set of the fourth passenger trajectory set, Detection set for the second passenger, is the passenger trajectory prediction value set of the fourth passenger trajectory set, is the ordinary IoU cost matrix, is the fourth calculation coefficient.

[0031] Preferably, the calculating the association cost between the third passenger detection set and the target set includes:

[0032]

[0033] Where, is the association cost between the third passenger detection set and the target set, For the third passenger detection collection, The target set.

[0034] Preferably, the preceding traces in the total set The unreliability of the passenger detection value associated with the trajectory in the current frame and the total set is calculated by the frame detection value, including:

[0035] Get the track based on the first The mean and standard deviation of the historical detection box height changes within the frame are used to calculate the threshold;

[0036]

[0037]

[0038]

[0039] Where, is the threshold, is the standard deviation, is the mean, is a natural number, is the height difference between adjacent detection boxes on the trajectory.

[0040] Preferably, it also includes:

[0041] The absolute value of the height difference between the current passenger detection frame and the last historical detection frame of its associated trajectory is calculated. If the absolute value of the height difference is greater than the threshold, the current passenger detection frame is considered an outlier and its unreliability is calculated using the sigmoid function.

[0042] Preferably, updating the predicted trajectory using the corrected passenger detection value includes:

[0043] Treat the fourth passenger trajectory set as a new trajectory and add it to the passenger trajectory set;

[0044] It is determined whether the unupdated time of the sixth passenger trajectory set is greater than the maximum survival time of the trajectory. If so, the trajectory is deleted from the passenger trajectory set.

[0045] The technical solution of the present invention has at least the following advantages and beneficial effects:

[0046] This method takes passengers in the terminal as the main tracking objects and combines efficient video surveillance data to achieve real-time multi-target tracking in the terminal scene.

[0047] The multi-target tracking method proposed in this paper can analyze terminal surveillance video in real time and accurately track multiple passengers. This technology not only improves monitoring efficiency, such as enabling passenger counting and early warning announcements, but also optimizes passenger flow, reduces congestion, and thus improves the overall passenger travel experience.

[0048] Combined with the current aviation industry's growing demand for safety and efficiency, this method will provide strong data support for airport management departments, enabling them to respond to emergencies in a timely manner based on real-time monitoring and improve the intelligence level of airport operations. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without paying any creative work.

[0050] Figure 1 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION

[0051] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of 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. Generally, the components of the embodiments of the present invention described and shown in the drawings herein can be arranged and designed in various different configurations.

[0052] The terms "first," "second," and so on, in the specification and claims of this application and the accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. The naming or numbering of steps in this application does not necessarily imply that the steps in the method flow must be executed in the chronological or logical order indicated by the naming or numbering. Named or numbered process steps may be executed in a different order based on the desired technical objectives, as long as the same or similar technical effects are achieved.

[0053] The module divisions described in this application are logical. In actual implementation, other divisions may be employed. For example, multiple modules may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the connection, coupling, or communication described in this application may be a direct connection, coupling, or communication between related objects, or an indirect connection, coupling, or communication through other devices. Furthermore, the connection, coupling, or communication between objects may be electrical or other similar forms, which are not limited in this application.

[0054] Please refer to Figure 1 , a real-time multi-target tracking method for passengers in an airport terminal, comprising:

[0055] S101: Obtain the confidence of the passenger detection set in the target video frame using the YOLOX detector and set a confidence threshold. The passenger detection set with a confidence level greater than the confidence threshold is classified as a first passenger detection set, and the passenger detection set with a confidence level less than the confidence threshold is classified as a second passenger detection set.

[0056] Specifically, airport terminal design encompasses multiple distinct scenarios, each with significant variations in elements. For example, boarding gates can be brightly lit, while security checkpoints can be dim. These factors can affect detector performance, leading to noise in target detection and significantly impacting subsequent tracking. Furthermore, target detection in terminal scenarios must meet real-time requirements. Therefore, we employ the YOLOX detector, which balances real-time performance with accuracy, for passenger target detection.

[0057] Input the current video frame into the YOLOX model to obtain the upper left corner coordinates, lower right corner coordinates and confidence of the detection frame of all passengers in the current video frame. Set the threshold of the high confidence detection frame , add the passenger detection boxes with confidence higher than the threshold to the first passenger detection set , add the passenger detection boxes with confidence lower than the threshold to the second passenger detection set .

[0058] S102: Obtain a set of historical passenger trajectories, obtain a set of passenger trajectory prediction values ​​based on a Kalman filter, establish an evaluation index based on the relationship between the speed and direction of the historical passenger trajectories and the height change of the trajectory detection box, establish an initial association cost based on the trajectory detection box and the evaluation index, and establish an association cost between the first passenger detection set and the passenger trajectory prediction value set based on the initial association cost to obtain a first cost matrix;

[0059] A new IoU is designed by considering the relationship between the speed direction of the passenger's historical trajectory and the height change of the detection box on the trajectory, and a new association cost is designed based on the constancy of the IoU between the trajectory detection boxes, so as to further improve the passenger detection value of the first passenger detection set. and passenger trajectory prediction value set The association costs between them are used to obtain a new cost matrix.

[0060] S103: Applying the Hungarian algorithm to the first cost matrix yields the first set of successfully matched passenger trajectories. , the second passenger trajectory set that does not match any detection and the third passenger detection set that does not match any trajectory ;

[0061] S104: Calculate the second passenger detection set Collected with the second passenger trajectory The second cost matrix of the passenger trajectory prediction value set is applied to the second cost matrix by the Hungarian algorithm to obtain the third passenger trajectory set that is successfully matched. and the fourth passenger trajectory set that does not match any detection ;

[0062] S105: Obtaining the fourth passenger's trajectory The last matching detection in the set constitutes the target set , calculate the third passenger detection set With target collection The third cost matrix is ​​obtained by applying the Hungarian algorithm to the third cost matrix, and the fifth set of passenger trajectories that are successfully matched is output. , the sixth passenger trajectory set that does not match any detection and the fourth passenger detection set that does not match any trajectory .

[0063] This method takes passengers in the terminal as the main tracking objects and combines efficient video surveillance data to achieve real-time multi-target tracking in the terminal scene.

[0064] The multi-target tracking method proposed in this paper can analyze terminal surveillance video in real time and accurately track multiple passengers. This technology not only improves monitoring efficiency, such as enabling passenger counting and early warning announcements, but also optimizes passenger flow, reduces congestion, and thus improves the overall passenger travel experience.

[0065] An exemplary embodiment of the present invention further includes:

[0066] S201: Obtain a combined total set of the first passenger trajectory set, the third passenger trajectory set, and the fifth passenger trajectory set, and use a YOLOX detector to detect height changes of historical trajectory detection frames to correct the trajectories in the total set;

[0067] S202: Calculate the unreliability of the passenger detection value associated with the current frame and the trajectory in the total set based on the detection values ​​of the previous k frames of the trajectory in the total set, and correct the passenger detection value based on the unreliability;

[0068] S203: Update the predicted trajectory using the corrected passenger detection value and output the final predicted trajectory.

[0069] For the successfully matched passenger trajectory set We designed a historical detection information correction strategy to update the trajectory. Specifically, we use the change information of the detection box height of the detector's historical trajectory to correct the trajectory, and calculate the unreliability of the passenger detection value associated with the current frame and these trajectories based on the detection values ​​of the previous k frames of these trajectories. , and then use the unreliability to correct the passenger detection value.

[0070] The trajectory is updated using the adjusted detection value to obtain a more accurate subsequent predicted trajectory.

[0071] In an exemplary embodiment of the present invention, obtaining a set of historical passenger trajectories and obtaining a set of passenger trajectory prediction values ​​based on a Kalman filter includes:

[0072] For the video frame at time t-1 in the surveillance video, it is assumed that there is already a trajectory set , from the set Take out a trajectory from , and then use Kalman filtering to predict the state prediction value at time t and the prediction error covariance matrix ,

[0073]

[0074]

[0075] Where, for The predicted state value at the moment, is the predicted change matrix, is a track in the historical passenger track set, is the forecast error covariance matrix, is the updated forecast error covariance matrix, is the transposed matrix of the predicted change matrix, is the noise matrix.

[0076] Preferably, establishing an evaluation index based on the relationship between the speed direction of the passenger's historical trajectory and the height change of the trajectory detection frame includes:

[0077]

[0078] Where, As the evaluation index, Prediction box height, is the height of the detection box.

[0079] Preferably, the association cost for establishing the first passenger detection set and the passenger trajectory prediction value set based on the initial association cost includes:

[0080] All tracks Composing passenger trajectory prediction value set , then and First, calculate and The association cost between them can be calculated as follows:

[0081]

[0082] Where, is the association cost between the first passenger detection set and the passenger trajectory prediction value set, For the first passenger detection collection, is the set of passenger trajectory prediction values, is the cost of the evaluation indicator, is the first weight parameter, is the cost of the orientation consistency between the detection box and the detection box on the trajectory, is the second weight parameter, is the detection box confidence cost, is the third weight parameter, is the initial association cost.

[0083] in, A new association cost is proposed for this scheme, which is an IoU cost that considers the relationship between the speed direction of the passenger's historical trajectory and the height change of the passenger detection box.

[0084] However, when the objects are highly overlapped, their distinguishing ability will become weaker, so we introduced VHIoU, which is calculated as follows:

[0085]

[0086] in, Indicates that the direction of the passenger's historical trajectory is towards the camera. The direction of the passenger's historical trajectory is away from the camera. Indicates that the direction of the passenger's historical trajectory is parallel to the camera. is the height of the passenger detection frame in the current frame, For the The height of the passenger detection box in the previous frame of the trajectory. and Used to adjust IoU, where is the threshold, Controls the magnitude of IoU scaling.

[0087] After the above adjustment of IoU, when the speed direction of the k-th passenger trajectory is toward the camera, the trajectory prediction box will be appropriately enlarged ( ) and the height is greater than the detection frame of the previous frame on the trajectory ( )’s detection box (denoted as ). For a detection frame whose height is smaller than the detection frame of the previous frame on the trajectory (denoted as ), reduce its On the contrary, when the velocity direction of the passenger trajectory is away from the camera, we zoom in and reduce the IoU This adjustment makes the adjusted IoU have a stronger ability to distinguish severely overlapping targets when the targets are severely overlapped, thus making the association more accurate.

[0088]

[0089] Where, is the confidence of the trajectory prediction box, is the confidence of the detection box.

[0090] is a new association cost proposed in this scheme, which further enhances the association performance by considering the IoU stability between the historical detection frames of passengers. Tracks and For each track, select the one in front of it. The historical detection frame sequence within the frame ( ). Then calculate the IoU value between adjacent detection frames to form a length of IoU sequence: , calculate the average value of the sequence, denoted as The calculation formula is as follows:

[0091]

[0092] By calculating the IoU between the new detection box and the passenger trajectory ( ) and the IoU of passenger trajectories ( The difference between them is calculated as follows:

[0093]

[0094] In getting 、 、 and After that, the weighted sum is performed to obtain the final association cost , 、 and is the weight parameter. Finally, the Hungarian algorithm is applied to the cost matrix to obtain the set of successfully matched passenger trajectories. , a set of passenger trajectories that do not match any detection , a set of passenger detections that do not match any trajectory .

[0095] Next, let The set of trajectory prediction values and , the associated cost is calculated as follows:

[0096] The second cost matrix of the passenger trajectory prediction value set of the second passenger detection set and the fourth passenger trajectory set includes:

[0097]

[0098] Where, is the association cost between the second passenger detection set and the passenger trajectory prediction value set of the fourth passenger trajectory set, Detection set for the second passenger, is the passenger trajectory prediction value set of the fourth passenger trajectory set, is the ordinary IoU cost matrix, is the fourth calculation coefficient.

[0099] The association cost here uses the common IoU cost and confidence cost. Similarly, the Hungarian algorithm is used on the cost matrix to obtain the successfully matched passenger trajectory. And update the set of passenger trajectories that do not match any detection to obtain .

[0100] Calculating the association cost between the third passenger detection set and the target set includes:

[0101]

[0102] Where, is the association cost between the third passenger detection set and the target set, For the third passenger detection collection, The target set.

[0103] In one embodiment, the preceding traces in the total set are The unreliability of the passenger detection value associated with the trajectory in the current frame and the total set is calculated by the frame detection value, including:

[0104] Get the track based on the first The mean and standard deviation of the historical detection box height changes within the frame are used to calculate the threshold;

[0105]

[0106]

[0107]

[0108] Where, is the threshold, is the standard deviation, is the mean, is a natural number, is the height difference between adjacent detection boxes on the trajectory.

[0109] Then calculate the height change between the current passenger detection frame and the last historical detection frame of its associated trajectory .if , we regard the detection box as an “outlier” and use the sigmoid function to calculate its unreliability , the calculation formula is as follows:

[0110]

[0111] In the updating process of the Kalman filter, for the Passenger trajectories, using passenger detection values Forecast value Update and get the updated status value :

[0112]

[0113] In this process, use Adjust passenger detection values , thereby effectively adjusting the detection value according to its unreliability. The specific formula is as follows:

[0114]

[0115] when When , the passenger detection value is completely reliable and does not require any adjustment. When the passenger detection value is completely unreliable, use the predicted value Alternative test value.

[0116] Where, is the observation matrix, which maps the predicted values ​​to the detection space.

[0117] Updating the predicted trajectory using the corrected passenger detection values ​​includes:

[0118] Treat the fourth passenger trajectory set as a new trajectory and add it to the passenger trajectory set;

[0119] It is determined whether the unupdated time of the sixth passenger trajectory set is greater than the maximum survival time of the trajectory. If so, the trajectory is deleted from the passenger trajectory set.

[0120] Specifically, after the passenger trajectory update phase is completed, for the passenger detection set that does not match any passenger trajectory , treat them as new trajectories and add them to the passenger trajectory set For passenger trajectory sets that do not match any detection , judge whether its unupdated time is greater than the maximum survival time of the trajectory. If it is greater than the maximum survival time of the trajectory, then the trajectory is removed from the passenger trajectory set Through the above process, we obtain the passenger trajectory set updated in the current video frame. These trajectory sets continue to repeat the prediction and update process as time changes, thereby realizing multi-target tracking of passengers in the terminal.

[0121] In addition, the functional units in the various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0122] If the integrated unit is implemented as a software functional unit and sold or used as a standalone product, it can be stored on a computer-readable storage medium. This computer software product, stored on a storage medium, includes instructions for causing a computer device (which may be a personal computer, server, or network device, etc.) to perform all or part of the steps of the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0123] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

Claims

1. A real-time multi-target tracking method for passengers in an airport terminal, characterized in that: include: The confidence level of the passenger detection set in the target video frame is obtained using the YOLOX detector, and a confidence threshold is set. The passenger detection set with a confidence level greater than the confidence threshold is divided into the first passenger detection set, and the passenger detection set with a confidence level less than the confidence threshold is divided into the second passenger detection set. Obtain a set of historical passenger trajectories, obtain a set of passenger trajectory prediction values ​​based on a Kalman filter, establish an evaluation index based on the relationship between the speed and direction of the historical passenger trajectories and the height change of the trajectory detection box, establish an initial association cost based on the trajectory detection box and the evaluation index, and establish an association cost between the first passenger detection set and the passenger trajectory prediction value set based on the initial association cost to obtain a first cost matrix; Applying the Hungarian algorithm to the first cost matrix yields a first set of successfully matched passenger trajectories, a second set of unmatched passenger trajectories, and a third set of unmatched passenger detections. Calculate a second cost matrix between the second passenger detection set and the set of passenger trajectory prediction values ​​for the second passenger trajectory set, and apply the Hungarian algorithm to the second cost matrix to obtain a third set of successfully matched passenger trajectories and a fourth set of passenger trajectories that do not match any detections. Obtain the last matching detection in the fourth passenger trajectory set to form the target set. Calculate the association cost between the third passenger detection set and the target set to obtain a third cost matrix. Apply the Hungarian algorithm to the third cost matrix and output the fifth passenger trajectory set that successfully matches, the sixth passenger trajectory set that does not match any detection, and the fourth passenger detection set that does not match any trajectory. The evaluation index established by the relationship between the speed direction of the passenger's historical trajectory and the height change of the trajectory detection frame includes: Where, As the evaluation index, Prediction box height, is the height of the detection frame; The association cost for establishing the first passenger detection set and the passenger trajectory prediction value set based on the initial association cost includes: Where, is the association cost between the first passenger detection set and the passenger trajectory prediction value set, For the first passenger detection collection, is the set of passenger trajectory prediction values, is the cost of the evaluation indicator, is the first weight parameter, is the cost of the orientation consistency between the detection box and the detection box on the trajectory, is the second weight parameter, is the detection box confidence cost, is the third weight parameter, is the initial association cost.

2. A method for real-time multi-target tracking of passengers in an airport terminal according to claim 1, characterized in that: Also includes: Obtain a combined total set of the first, third, and fifth passenger trajectory sets, and use the YOLOX detector to detect changes in the height of the historical trajectory detection frames to correct the trajectories in the total set; The unreliability of the passenger detection value associated with the trajectory in the total set is calculated based on the detection values ​​of the previous k frames of the trajectory in the total set, and the passenger detection value is corrected based on the unreliability. The predicted trajectory is updated using the corrected passenger detection value, and the final predicted trajectory is output.

3. The method for real-time multi-target tracking of passengers in an airport terminal according to claim 1, characterized in that: The acquisition of a historical passenger trajectory set and the acquisition of a passenger trajectory prediction value set based on a Kalman filter include: Where, for The predicted state value at the moment, is the predicted change matrix, is a track in the historical passenger track set, is the forecast error covariance matrix, is the updated forecast error covariance matrix, is the transposed matrix of the predicted change matrix, is the noise matrix.

4. A method for real-time multi-target tracking of passengers in an airport terminal according to claim 3, characterized in that: The second cost matrix of the passenger trajectory prediction value set of the second passenger detection set and the fourth passenger trajectory set includes: Where, is the association cost between the second passenger detection set and the passenger trajectory prediction value set of the fourth passenger trajectory set, Detection set for the second passenger, is the passenger trajectory prediction value set of the fourth passenger trajectory set, is the ordinary IoU cost matrix, is the fourth calculation coefficient.

5. A method for real-time multi-target tracking of passengers in an airport terminal according to claim 4, characterized in that: Calculating the association cost between the third passenger detection set and the target set includes: Where, is the association cost between the third passenger detection set and the target set, For the third passenger detection collection, The target set.

6. The method for real-time multi-target tracking of passengers in an airport terminal according to claim 2, characterized in that: The front of the trajectory in the total set The unreliability of the passenger detection value associated with the trajectory in the current frame and the total set is calculated by the frame detection value, including: Get the track based on the first The mean and standard deviation of the historical detection box height changes within the frame are used to calculate the threshold; Where, is the threshold, is the standard deviation, is the mean, is a natural number, is the height difference between adjacent detection boxes on the trajectory.

7. A method for real-time multi-target tracking of passengers in an airport terminal according to claim 6, characterized in that: Also includes: The absolute value of the height difference between the current passenger detection frame and the last historical detection frame of its associated trajectory is calculated. If the absolute value of the height difference is greater than the threshold, the current passenger detection frame is considered an outlier and its unreliability is calculated using the sigmoid function.

8. The method for real-time multi-target passenger tracking in an airport terminal according to claim 2, characterized in that: The updating of the predicted trajectory using the corrected passenger detection value includes: Treat the fourth passenger trajectory set as a new trajectory and add it to the passenger trajectory set; It is determined whether the unupdated time of the sixth passenger trajectory set is greater than the maximum survival time of the trajectory. If so, the trajectory is deleted from the passenger trajectory set.

Images