Target tracking optimization method and system based on historical trajectory prediction of holder camera

Abstract

The invention discloses a target tracking optimization method and system based on pan-tilt camera historical trajectory prediction. The method comprises the steps of obtaining a target angle trajectory; predicting target angles of multiple frames in the future through the target angle track; predicting a future multi-frame holder orientation angle through a preset instruction sequence; converting the target angles of the plurality of future frames and the holder orientation angle difference values of the plurality of future frames into target position coordinates in a picture, and generating a target position sequence of the plurality of future frames; performing weighted summation on the target position of the previous frame and the predicted position of the current frame, adjusting a search center point of a tracking algorithm, and outputting the target position of the current frame; and according to the total delay of the system, performing time sequence compensation on the current target position and the future predicted position, and generating a reference position to offset the hysteresis effect. According to the method, the target trajectory is reconstructed and predicted by using the angle information returned by the pan-tilt camera, so that the success rate of the target tracking algorithm is improved, and the hysteresis effect caused by system delay is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of single target tracking of computer vision, and in particular to a target tracking optimization method and system based on historical trajectory prediction of a pan-tilt camera. Background Art

[0002] A pan-tilt camera is a camera device that integrates horizontal and vertical rotation and optical zoom functions. It can rotate and zoom through commands, and dynamically adjust the viewing angle and field of view. Compared with fixed cameras, pan-tilt cameras can not only cover a wider monitoring area, but also track key targets through zoom to obtain high-resolution detail information. It has important application value in the fields of intelligent security monitoring (such as dynamic tracking of suspicious targets), vehicle-road collaborative systems (real-time generation of dynamic traffic maps), and drone airspace control (precisely locating illegal aircraft and linking countermeasures). The core technical challenge lies in the coordinated optimization of real-time target tracking accuracy and pan-tilt mechanical response speed in complex scenarios.

[0003] The existing PTZ camera tracking system usually consists of two parts: the perception module and the control module. A common implementation of the perception module is the YOLO series target detection algorithm combined with target matching or simple motion model Kalman filtering. Another implementation is to first select the target through the target detection algorithm and then track it through the single target tracking algorithm of the Siamese network series.

[0004] The first implementation completely relies on the target detection results. It is easy to track other targets when interfered by similar targets. When the target is occluded, it will cause greater interference to the Kalman filter results. In addition, the simple motion model Kalman filter is difficult to adapt to the nonlinear motion of the target, such as sudden stop and change of direction. The target category needs to be an existing category in the detection algorithm model, and it is impossible to detect and track unknown target types.

[0005] The second implementation usually has a higher computational complexity. Compared with the field of target detection, there are fewer studies in the field of target tracking, and there are relatively few practical algorithm implementations and optimizations, which are more difficult to deploy in actual scenarios; target tracking usually relies on the small motion assumption (assuming that the position of the target does not change much between two adjacent frames) to reduce the target search range and thus improve the real-time performance of the algorithm, but it is easy to lose tracking when facing a large rotation of the camera's view or violent movement of the target; target tracking can track any category of targets, but usually does not have a detection function, and requires the detection algorithm to provide a target template image to initialize the tracking algorithm.

[0006] The control module usually uses PID closed-loop feedback to achieve tracking, and its working process is as follows: taking the horizontal and vertical pixel offsets between the center point of the target circumscribed rectangle output by the tracking algorithm and the center point of the image as the rotation error signal, calculating the pan-tilt rotation angle command to drive the camera to aim at the target; at the same time, taking the difference between the pixel size of the target circumscribed box and the preset expected size as the zoom error signal, calculating the optical zoom multiple of the camera to keep the target at the best display ratio in the picture. Finally, a closed-loop control loop of "target tracking - error calculation - command generation - mechanical response" is formed to ensure that the target continuously and stably stays in the center area of the field of view.

[0007] The latency of the pan-tilt camera tracking system is mainly composed of camera acquisition latency, algorithm latency, and mechanical response latency. The total latency is usually within dozens of milliseconds to hundreds of milliseconds, resulting in a lag in view adjustment. The cumulative effect of this lag causes significant offsets in the positions of the target in adjacent frames, violating the "small motion assumption" relied on by the tracking algorithm and making it easy to lose the tracking.

[0008] The Chinese invention patent application with the application number: 202311465351.5 discloses "A Multi-Target Visual Tracking Method Based on Camera Motion Trend Estimation", including: obtaining video stream data, and obtaining the pixel coordinates of the target in the image plane through the target detection result; obtaining camera parameters, and using homography transformation to map the target in the image plane to the ground plane to obtain the mapped measurement value; calculating the corresponding process compensation factor according to the motion trend of the camera, and then calculating the process noise covariance matrix; calculating the measurement distribution of the target on the ground plane, feeding this distribution back into the Kalman filter for iteration, and correcting the Kalman filter according to the process noise covariance matrix. The Kalman filter predicts and outputs the trajectory state of the target; using the mapped measurement value and the predicted trajectory state as inputs to calculate the mapped Mahalanobis distance, and storing the result in the association cost matrix; according to the association cost matrix, associating the effective detection boxes with the predicted trajectories through the Hungarian algorithm to improve the tracking robustness and accuracy. When the motion of the target is non-linear motion, this technical solution has great limitations. Summary of the Invention

[0009] To solve the technical problem of the lag effect caused by system latency in the existing target tracking system, the present invention provides an optimized method and system for target tracking based on the prediction of the historical trajectory of a pan-tilt camera. The technical solution adopted by the present invention is:

[0010] The first aspect of the present invention provides an optimized method for target tracking based on the prediction of the historical trajectory of a pan-tilt camera, and the method includes:

[0011] Obtain the target angle trajectory;

[0012] Predict the target angles of multiple future frames through the target angle trajectory;

[0013] Predict the pan-tilt orientation angles for multiple future frames through a preset instruction sequence;

[0014] Convert the difference between the target angles of the multiple future frames and the pan-tilt orientation angles of the multiple future frames into target position coordinates in the picture, and generate a target position sequence for the multiple future frames;

[0015] Weighted sum the target position of the previous frame and the predicted position of the current frame, adjust the search center point of the tracking algorithm, and output the target position of the current frame;

[0016] According to the total system delay, perform timing compensation on the current target position and future predicted positions to generate reference positions to offset the lag effect.

[0017] As a preferred solution, the method for obtaining the target angle trajectory includes:

[0018] Obtain the azimuth and elevation angle trajectories of the target: According to the azimuth φ cam and elevation angle θ cam transmitted in real time by the pan-tilt camera, combined with the coordinates (x, y) of the target center relative to the picture center, the horizontal field of view angle F h of the camera, the vertical field of view angle F v of the camera, the picture width W and height H, calculate the offsets Δφ tgt and Δθ tgt of the target relative to the pan-tilt orientation through the frustum model:

[0019]

[0020] Respectively add the azimuth φ cam and elevation angle θ cam of the current pan-tilt camera orientation, to obtain the target azimuth φ tft and elevation angle θ tgt :

[0021] φ tgt = φ cam + Δφ tgt

[0022] θ tgt = θ tgt + Δθ tgt

[0023] Perform the above conversion and recording for each frame, and obtain a target angle trajectory at time t:

[0024]

[0025] As a preferred solution, the method for predicting the target angles of multiple future frames through the target angle trajectory includes:

[0026] Using a sequence-to-sequence deep learning model, based on the latest M frames of data in the target angle trajectory, autoregressively generate the predicted results of the target angle for the next K frames

[0027]

[0028] As a preferred solution, the method for predicting the pan-tilt orientation angles of multiple future frames through a preset instruction sequence includes:

[0029] Using a sequence-to-sequence deep learning model, based on the latest M frames of instructions in the instruction sequence, autoregressively generate the predicted results of the pan-tilt orientation angles of the next K-1 frames of multiple future frames

[0030]

[0031] As a preferred solution, the method for converting the difference between the target angles of the multiple future frames and the pan-tilt orientation angles of the multiple future frames into the target position coordinates in the picture and generating the target position sequence of the multiple future frames includes:

[0032] Subtract the pan-tilt orientation angle from the predicted target angle to obtain the angle of the target relative to the pan-tilt orientation:

[0033]

[0034] Convert the angle back to the target position in the picture:

[0035]

[0036] Perform the above conversion on the next K frames to obtain the target position estimation of the next K frames

[0037]

[0038] As a preferred solution, the method for weighted summation of the previous frame target position and the current frame predicted position to adjust the search center point of the tracking algorithm and output the current frame target position includes:

[0039] Use the target center position output by the tracking algorithm in the previous frame And the predicted target position of the first frame Perform weighted summation to obtain a reference search position:

[0040]

[0041] where λ∈[0,1] represents the degree of trust in the prediction result;

[0042] Subsequently, the search center point of the tracking algorithm is changed from the previous frame target position P t-1Change to the reference search position P alg Run the tracking algorithm to obtain the target center position P of the current frame t .

[0043] As a preferred solution, according to the total system delay, the method of performing timing compensation on the current target position and future predicted positions to generate a reference position to offset the lag effect includes:

[0044] The average delay from when the camera starts to capture a frame of video to when the mechanical response is completed is:

[0045] τ = τ cam + τ alg + τ rsp

[0046] where τ cam is the average delay of camera capture, τ alg is the average delay of the tracking algorithm, and τ rsp is the average delay of mechanical response;

[0047] That is, the control algorithm actually tracks the target τ ago, corresponding to the target frames ago, where Δt is the system processing interval;

[0048] Perform a weighted calculation on the current frame target position P t obtained by the tracking algorithm and the predicted target positions of the subsequent d - 1 frames to calculate the reference target position:

[0049]

[0050] where, represents P t , σ is the attenuation factor, and the weight decays exponentially with the prediction step;

[0051] Then use the reference target position P ctl to replace P t to calculate the position error signal.

[0052] The second aspect of the present invention provides an optimized target tracking system based on the historical trajectory prediction of a pan - tilt camera. The system includes an angle trajectory acquisition module, a target angle prediction module, a pan - tilt orientation angle prediction module, a target position calculation module, a tracking algorithm optimization module, and a control algorithm optimization module;

[0053] The angle trajectory acquisition module is used to acquire the target angle trajectory;

[0054] The target angle prediction module is used to predict the target angles of multiple future frames through the target angle trajectory;

[0055] The pan-tilt orientation angle prediction module is used to predict the pan-tilt orientation angles of multiple future frames through a preset instruction sequence;

[0056] The target position calculation module is used to convert the difference between the target angles of the multiple future frames and the pan-tilt orientation angles of the multiple future frames into target position coordinates in the picture, and generate a target position sequence of multiple future frames;

[0057] The tracking algorithm optimization module is used to perform weighted summation of the previous frame target position and the current frame predicted position, adjust the search center point of the tracking algorithm, and output the current frame target position;

[0058] The control algorithm optimization module is used to perform timing compensation on the current target position and future predicted positions according to the total system delay, and generate a reference position to offset the lag effect.

[0059] The third aspect of the present invention provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the foregoing target tracking optimization method based on the historical trajectory prediction of a pan-tilt camera are implemented.

[0060] The fourth aspect of the present invention provides a computer device, including a storage medium, a processor, and a computer program stored in the storage medium and executable by the processor. When the computer program is executed by the processor, the steps of the foregoing target tracking optimization method based on the historical trajectory prediction of a pan-tilt camera are implemented.

[0061] Compared with the prior art, the beneficial effects of the present invention are:

[0062] The present invention reconstructs the target trajectory by combining the azimuth angle and elevation angle information transmitted back by the pan-tilt camera with the position information of the target in the picture, and predicts the future angle of the target through a sequence-to-sequence deep learning model. Through the sequence-to-sequence deep learning model, the historical instruction sequence of the pan-tilt is input to predict the future state (azimuth angle, elevation angle) of the pan-tilt. The future position of the target on the picture is calculated by combining the predicted future angle of the pan-tilt with the future angle of the target (integrating the motion information of the pan-tilt and the motion of the target).

[0063] The search range of the tracking algorithm is corrected by the future position of the target on the picture, improving the success rate of the tracking algorithm

[0064] Combining a series of predicted future positions of the target and the current position of the target, performing weighted summation to generate a position error signal to offset the system delay. The weights are allocated according to the measured system delay. Description of the Drawings

[0065] Figure 1 It is a flowchart of the target tracking optimization method based on the historical trajectory prediction of a pan-tilt camera provided in this embodiment. Detailed implementation manners

[0066] The accompanying drawings are only for illustrative purposes and should not be construed as limiting the present invention;

[0067] It should be clear that the described embodiments are only a part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope protected by the embodiments of the present application.

[0068] The terms used in the embodiments of the present application are only for the purpose of describing specific embodiments, and are not intended to limit the embodiments of the present application. The singular forms of "a", "the" and "said" used in the embodiments of the present application and the appended claims are also intended to include the plural forms, unless the context clearly indicates otherwise. It should also be understood that the term " / and / " as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0069] When the following description refers to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present application. On the contrary, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims. In the description of the present application, it should be understood that the terms "first", "second", "third", etc. are only used to distinguish similar objects, and do not have to be used to describe a specific order or sequence, nor can they be understood as indicating or implying relative importance. For those of ordinary skill in the art, the specific meanings of the above terms in the present application can be understood according to specific circumstances.

[0070] In addition, in the description of the present application, unless otherwise specified, "a plurality of" means two or more. " / and / " describes the association relationship of associated objects, indicating that three relationships can exist. For example, A and / or B can represent: A exists alone, A and B exist simultaneously, and B exists alone. The character " / " generally represents an "or" relationship between the associated objects before and after. The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0071] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0072] Embodiment 1

[0073] Please refer to Figure 1 , this embodiment provides an optimized method for target tracking based on the historical trajectory prediction of a pan-tilt camera, and the method includes:

[0074] S1: Obtain the target angle trajectory;

[0075] In a specific embodiment, the method for obtaining the target angle trajectory includes:

[0076] Obtain the azimuth and elevation angle trajectories of the target: According to the azimuth angle φ cam and elevation angle θ cam transmitted in real time by the pan-tilt camera, combined with the coordinates (x, y) of the target center relative to the center of the screen (the coordinate origin is the center of the screen, with the right and upward directions being positive), the horizontal field of view angle F h of the camera, the vertical field of view angle F v , the width W and height H of the screen, calculate the offsets Δφ tgt and Δθ tgt of the target relative to the orientation of the pan-tilt through the frustum model:

[0077]

[0078] Respectively add the azimuth angle φ cam and elevation angle θ cam of the current pan-tilt camera orientation, to obtain the target azimuth angle φ tgt and elevation angle θ tgt :

[0079] φ tgt = φ cam + Δφ tgt

[0080] θ tgt = θ tgt + Δθ tgt

[0081] Perform the above conversion and record for each frame, to obtain a target angle trajectory at time t:

[0082]

[0083] Specifically, the pan-tilt camera is: a 1080P PTZ camera, supporting horizontal rotation and elevation rotation.

[0084] S2: Predict the target angles of future multiple frames through the target angle trajectory;

[0085] In a specific embodiment, the method for predicting the target angles of future multiple frames through the target angle trajectory includes:

[0086] Adopt a sequence-to-sequence deep learning model (Transformer encoder-decoder model), and based on the latest M frames of data in the target angle trajectory, autoregressively generate the target angle prediction results of the future K frames (including the current frame)

[0087]

[0088] S3: Predict the pan-tilt orientation angles for multiple future frames through a preset instruction sequence;

[0089] In a specific embodiment, the method for predicting the pan-tilt orientation angles for multiple future frames through a preset instruction sequence includes:

[0090] Adopt a sequence-to-sequence deep learning model (Transformer encoder-decoder model), and based on the latest M frames of instructions in the instruction sequence, autoregressively generate the prediction results of the pan-tilt orientation angles for multiple future frames of the next K - 1 frames (excluding the current frame, and the pan-tilt state of the current frame has been obtained through the pan-tilt feedback information)

[0091]

[0092] S4: Convert the difference between the target angles for the multiple future frames and the pan-tilt orientation angles for the multiple future frames into target position coordinates in the picture, and generate a target position sequence for multiple future frames;

[0093] In a specific embodiment, the method for converting the difference between the target angles for the multiple future frames and the pan-tilt orientation angles for the multiple future frames into target position coordinates in the picture and generating a target position sequence for multiple future frames includes:

[0094] Subtract the pan-tilt orientation angle from the predicted target angle to obtain the angle of the target relative to the pan-tilt orientation:

[0095]

[0096] Convert the angle back to the target position in the picture:

[0097]

[0098] Perform the above conversion on the next K frames (including the current frame) to obtain the target position estimates for the next K frames

[0099] S5: Weightedly sum the target position of the previous frame and the predicted position of the current frame, adjust the search center point of the tracking algorithm (SiamRPN), and output the target position of the current frame;

[0100] In a specific embodiment, the method for weightedly summing the target position of the previous frame and the predicted position of the current frame, adjusting the search center point of the tracking algorithm, and outputting the target position of the current frame includes:

[0101] Use the target center position output by the tracking algorithm in the previous frame And the predicted target position of the first frame (current frame) Perform weighted summation to obtain a reference search position:

[0102]

[0103] where λ ∈ [0, 1] represents the degree of confidence in the prediction result;

[0104] Subsequently, change the search center point of the tracking algorithm from the target position P in the previous frame t-1 to the reference search position P alg , run the tracking algorithm, and obtain the target center position P in the current frame t .

[0105] S6: According to the total system delay, perform temporal compensation on the current target position and future predicted positions to generate a reference position to offset the lag effect;

[0106] In a specific embodiment, the method for performing temporal compensation on the current target position and future predicted positions according to the total system delay to generate a reference position to offset the lag effect includes:

[0107] The average delay from when the camera starts collecting a frame of video to when the mechanical response is completed is:

[0108] τ = τ cam + τ alg + τ rsp

[0109] where τ cam is the average camera acquisition delay, τ alg is the average tracking algorithm delay, and τ rsp is the average mechanical response delay;

[0110] That is, the control algorithm actually tracks the target τ ago, corresponding to the target Δt frames ago, where Δt is the system processing interval (assuming all modules have the same frequency);

[0111] Perform weighted calculation on the current frame target position P t obtained by the tracking algorithm and the predicted target positions of the subsequent d - 1 frames to calculate the reference target position:

[0112]

[0113] where, represents P t , σ is the attenuation factor, and the weight decays exponentially with the prediction step;

[0114] Then use the reference target position P ctl to replace P t to calculate the position error signal.

[0115] Example 2

[0116] This embodiment provides an object tracking optimization system based on the historical trajectory prediction of a pan-tilt camera. The system includes an angle trajectory acquisition module, an object angle prediction module, a pan-tilt orientation angle prediction module, an object position calculation module, a tracking algorithm optimization module, and a control algorithm optimization module;

[0117] The angle trajectory acquisition module is used to acquire the object angle trajectory;

[0118] The object angle prediction module is used to predict the object angles of future multiple frames through the object angle trajectory;

[0119] The pan-tilt orientation angle prediction module is used to predict the pan-tilt orientation angles of future multiple frames through a preset instruction sequence;

[0120] The object position calculation module is used to convert the difference between the object angles of the future multiple frames and the pan-tilt orientation angles of the future multiple frames into the object position coordinates in the picture, and generate an object position sequence of the future multiple frames;

[0121] The tracking algorithm optimization module is used to perform weighted summation of the previous frame's object position and the current frame's predicted position, adjust the search center point of the tracking algorithm, and output the current frame's object position;

[0122] The control algorithm optimization module is used to perform timing compensation on the current object position and future predicted positions according to the total system delay, and generate a reference position to offset the lag effect.

[0123] Example 3

[0124] A computer-readable storage medium stores a computer program thereon. When the computer program is executed by a processor, the steps of the object tracking optimization method based on the historical trajectory prediction of a pan-tilt camera described in Example 1 are implemented.

[0125] Example 4

[0126] A computer device includes a storage medium, a processor, and a computer program stored in the storage medium and executable by the processor. When the computer program is executed by the processor, the steps of the object tracking optimization method based on the historical trajectory prediction of a pan-tilt camera described in Example 1 are implemented.

[0127] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, rather than limitations on the implementation manners of the present invention. For those of ordinary skill 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 enumerate all implementation manners here. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the claims of the present invention.

Claims

1. An object tracking optimization method based on the prediction of the historical trajectory of a pan-tilt camera, characterized in that, The method includes: Obtaining a target angle trajectory; Predicting the target angles of multiple future frames through the target angle trajectory; Predicting the pan-tilt orientation angles of multiple future frames through a preset instruction sequence; Converting the difference between the target angles of the multiple future frames and the pan-tilt orientation angles of the multiple future frames into target position coordinates in the picture to generate a target position sequence of multiple future frames; Performing weighted summation of the previous frame target position and the current frame predicted position to adjust the search center point of the tracking algorithm and output the current frame target position; Performing timing compensation on the current target position and future predicted positions according to the total system delay to generate a reference position to offset the lag effect.

2. The target tracking optimization method based on the historical trajectory prediction of a pan-tilt camera according to claim 1, wherein The method for obtaining the target angle trajectory includes: Obtain the azimuth and elevation angle trajectories of the target: Based on the azimuth angle φ cam and elevation angle θ cam transmitted in real time by the pan-tilt camera, combined with the coordinates (x, y) of the target center relative to the center of the image, the horizontal field of view angle F h of the camera, the vertical field of view angle F v , the width W and height H of the image, calculate the offsets Δφ tgt and Δθ tgt of the target relative to the orientation of the pan-tilt through the frustum model: Add the azimuth angle φ of the current pan-tilt camera orientation respectively cam and the pitch angle θ cam , to obtain the target azimuth angle φ tgt and the pitch angle θ tgt : φ tgt = φ cam + Δφ tgt θ tgt = θ tgt + Δθ tgt Performing the above conversion on each frame and recording, and obtaining a target angle trajectory at time t:

3. The target tracking optimization method based on the historical trajectory prediction of a pan-tilt camera according to claim 2, characterized in that, The method for predicting the target angles of multiple future frames through the target angle trajectory includes: Using a sequence-to-sequence deep learning model, autoregressively generate the target angle prediction results for the next K frames based on the most recent M frames of data in the target angle trajectory 4. The target tracking optimization method based on the historical trajectory prediction of a pan-tilt camera according to claim 3, characterized in that, The method for predicting the pan-tilt orientation angles of multiple future frames through a preset instruction sequence includes: Using a sequence-to-sequence deep learning model, based on the most recent M frames of instructions in the instruction sequence, autoregressively generate the predicted results of the future gimbal orientation angles for the next K-1 frames 5. The target tracking optimization method based on the historical trajectory prediction of a pan-tilt camera according to claim 4, wherein The method for converting the difference between the target angles of the multiple future frames and the pan-tilt orientation angles of the multiple future frames into target position coordinates in the picture to generate a target position sequence of multiple future frames includes: Subtracting the pan-tilt orientation angle from the predicted target angle to obtain the angle of the target relative to the pan-tilt orientation: Converting the angle back to the target position in the picture: Perform the above conversion on the future K frames to obtain the estimated target positions of the future K frames 6. The target tracking optimization method based on the historical trajectory prediction of a pan-tilt camera according to claim 5, wherein The method for performing weighted summation of the previous frame target position and the current frame predicted position to adjust the search center point of the tracking algorithm and output the current frame target position includes: The target center position output by the tracking algorithm in the previous frame and the predicted target position in the first frame are weighted and summed to obtain a reference search position: where λ ∈ [0, 1] represents the degree of trust in the prediction result; Subsequently, the search center point of the tracking algorithm is changed from the target position P in the previous frame t-1 to the reference search position P alg , and the tracking algorithm is run to obtain the target center position P in the current frame t .

7. The target tracking optimization method based on the prediction of the historical trajectory of a pan-tilt camera according to claim 1, characterized in that The method for performing timing compensation on the current target position and future predicted positions according to the total system delay to generate a reference position to offset the lag effect includes: The average delay from when the camera starts to capture a frame of picture to when the mechanical response is completed is: τ = τ cam + τ alg + τ rsp Among them, τ cam is the average delay of camera acquisition, τ alg is the average delay of the tracking algorithm, and τ rsp is the average delay of mechanical response; That is, the control algorithm actually tracks the target before τ, corresponding to the target before the frame, where Δt is the system processing interval; For the current frame target position P obtained by the tracking algorithm t And the predicted target positions of the subsequent d - 1 frames Perform weighted calculation to obtain the reference target position: Among them, represents P t , where σ is the attenuation factor and the weight decays exponentially with the prediction step; Then use the reference target position P ctl to replace P t and calculate the position error signal.

8. An object tracking optimization system based on the prediction of the historical trajectory of a pan-tilt camera, characterized in that, The system includes an angle trajectory acquisition module, a target angle prediction module, a pan-tilt orientation angle prediction module, a target position calculation module, a tracking algorithm optimization module, and a control algorithm optimization module; The angle trajectory acquisition module is used to obtain a target angle trajectory; The target angle prediction module is used to predict the target angles of multiple future frames through the target angle trajectory; The pan-tilt orientation angle prediction module is used to predict the pan-tilt orientation angles of multiple future frames through a preset instruction sequence; The target position calculation module is used to convert the difference between the target angles of the multiple future frames and the pan-tilt orientation angles of the multiple future frames into target position coordinates in the picture to generate a target position sequence of multiple future frames; The tracking algorithm optimization module is used to perform weighted summation of the previous frame target position and the current frame predicted position to adjust the search center point of the tracking algorithm and output the current frame target position; The control algorithm optimization module is used to perform timing compensation on the current target position and future predicted positions according to the total system delay to generate a reference position to offset the lag effect.

9. A computer-readable storage medium storing a computer program thereon, characterized in that: When the computer program is executed by a processor, it implements the steps of the target tracking optimization method based on the historical trajectory prediction of a pan-tilt camera as described in any one of claims 1 to 7.

10. A computer device, characterized in that: It includes a storage medium, a processor, and a computer program stored in the storage medium and executable by the processor. When the computer program is executed by the processor, it implements the steps of the target tracking optimization method based on the historical trajectory prediction of the pan-tilt camera as described in any one of claims 1 to 7.

Images