Unmanned aerial vehicle autonomous tracking trajectory planning method
By using an autonomous tracking trajectory planning method for UAVs, the problems of tracking misjudgment and motion recognition under multiple highly similar targets were solved, and accurate path planning and target identification were achieved.
Patent Information
- Application Number
- CN202511023690.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-24
- Publication Date
- 2025-10-17
AI Technical Summary
During tracking, drones are prone to misjudging targets due to local similarities in adjacent images, leading to target confusion. Furthermore, existing technologies fail to accurately identify the target's movement patterns, affecting the reliability and effectiveness of tracking path planning.
By analyzing the feature similarity of the current and historical images of the drone, multiple tracking targets with high feature similarity are determined. Through kinematic models and data analysis, typical or atypical movement modes are identified, the target position is predicted, the primary tracking target is selected, and the path is planned.
It effectively avoids target confusion, improves the accuracy of tracking path planning and target identification capabilities, and is suitable for target differentiation in dynamic scenarios.
Smart Images
Figure CN120802988A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of unmanned aerial vehicles (UAVs), and in particular relates to a method for planning an autonomous tracking trajectory of an UAV. Background Art
[0002] During actual drone tracking, it's common for multiple targets to appear with highly similar features in adjacent single-frame images. Traditional tracking algorithms often rely solely on current image features for target matching, lacking comprehensive consideration of the target's motion characteristics. This can lead to misjudgments when faced with multiple highly similar targets due to local similarities in adjacent images, mistaking the interfering target for the true target, leading to target confusion. Once this happens, the drone's subsequent tracking path planning deviates from the correct direction, making it unable to continuously and stably track the true target, severely impacting the reliability and effectiveness of the drone's mission.
[0003] Furthermore, existing technologies have significant deficiencies in their ability to identify the motion patterns of tracked targets. Most tracking algorithms simplify target motion into a simple uniform linear motion model, failing to fully account for the complex motion patterns that targets can exhibit in real-world scenarios. This lack of accurate recognition of target motion makes it difficult for drones to accurately predict their trajectories and determine their positions based on the target's actual motion characteristics. This not only reduces target identification accuracy, making it easy for drones to lose their way among multiple targets, but also results in a lack of targeted tracking path planning.
[0004] To this end, the present invention provides a method for autonomous tracking trajectory planning of a UAV. Summary of the Invention
[0005] In order to make up for the deficiencies of the prior art, at least one technical problem raised in the background technology is solved.
[0006] The technical solution adopted by the present invention to solve the technical problem is: a method for autonomous tracking trajectory planning of a UAV, comprising the following steps: Based on the images taken by the drone during the current autonomous tracking and the images taken during the previous autonomous tracking, the feature similarity analysis of the tracking target is performed to determine whether there are multiple tracking targets with high feature similarity in the current autonomous tracking of the drone. If it appears, the movement mode of the tracking target is determined by analyzing the running data of the tracking target at multiple historical moments, where the movement mode includes typical movement and atypical movement; Determine the predicted movement position of the target at the current autonomous tracking moment based on the target's movement pattern, and compare and analyze it with the positions of multiple similar targets. Determine the primary target among the similar targets. The predicted motion position of the primary tracking target at the next unmanned aerial vehicle autonomous tracking time is determined according to the motion mode of the primary tracking target, and a tracking path is determined in combination with the actual position of the current unmanned aerial vehicle.
[0007] Further, the tracking target feature similarity analysis process comprises the following steps: The image feature extraction algorithm is used to perform feature processing on the image captured by the unmanned aerial vehicle in the current autonomous tracking and the image captured by the unmanned aerial vehicle in the historical previous time autonomous tracking, to obtain a comparison feature descriptor set and a reference feature descriptor set. The average similarity measure value between the reference feature descriptor set and the comparison feature descriptor set is calculated by using the Euclidean distance method, and if the average similarity measure value is greater than or equal to the average similarity measure threshold, the tracking target in the image captured by the unmanned aerial vehicle in the current autonomous tracking is marked as a tracking similar target. If the number of tracking similar targets is greater than 1, it indicates that there are multiple tracking targets with high feature similarity in the current unmanned aerial vehicle autonomous tracking.
[0008] Further, the comparison feature descriptor set and the reference feature descriptor set are obtained in the following manner: The image feature extraction algorithm is used to extract feature points of the tracking target in the image captured by the unmanned aerial vehicle in the historical previous time autonomous tracking, and a feature descriptor is generated for each extracted feature point to form a reference feature descriptor set. The image feature extraction algorithm is used to extract feature points of the target in the image captured by the unmanned aerial vehicle in the current autonomous tracking, and a feature descriptor is generated for each extracted feature point to form a comparison feature descriptor set.
[0009] Further, the process of determining the motion mode of the tracking target comprises the following steps: According to the typical motion model established by the kinematic equation, the absolute error between the predicted position and the actual position of the tracking target at each historical time is calculated to obtain a position error value. If the position error value is not within the preset position error range, the non-typical motion time of the tracking target is determined. Through the processing and analysis of the non-typical motion time, a non-typical motion maintenance value and a non-typical motion degree value are obtained, and the sum of the two values is obtained. The motion mode of the tracking target is determined by the non-typical motion performance value. If the motion mode of the tracking target matches any one of the typical motion models, the motion mode of the tracking target is typical motion. If the motion mode of the tracking target does not match all the typical motion models, the motion mode of the tracking target is non-typical motion.
[0010] Further, the specific acquisition manner of the position error value is: By using the collected actual position, speed and acceleration data of the tracking target at the historical multiple moments, the least square method or other optimization algorithm is used to determine the typical motion model parameters, to calculate the predicted position of the tracking target output by the typical motion model at each historical moment, so as to calculate the absolute error between the predicted position and the actual position of the tracking target at each historical moment, and obtain the position error value.
[0011] Further, the acquisition manner of the atypical motion maintenance value is: The proportion of the time length of the atypical motion moment in the historical moment is counted to obtain the atypical motion maintenance value of the tracking target. The acquisition manner of the atypical motion degree value is: Based on the atypical running moment, the absolute difference proportion of the position error value at the atypical running moment and the adjacent nearest preset position error range endpoint value is calculated to obtain the atypical running moment position error out-of-range ratio, and the atypical running moment position error out-of-range ratio is averaged to obtain the atypical motion degree value of the tracking target.
[0012] Further, the process of determining the predicted motion position of the tracking target at the current unmanned aerial vehicle autonomous tracking moment according to the motion mode of the tracking target is: If the motion mode of the tracking target is typical motion, the current unmanned aerial vehicle autonomous tracking moment is taken as input, and the predicted motion position of the tracking target is output by the typical motion model matched with the motion mode of the tracking target. If the motion mode of the tracking target is atypical motion, the motion trajectory of the tracking target at multiple historical moments is drawn according to the actual position of the tracking target at the historical multiple moments, and the motion trajectory of the tracking target at adjacent historical moments is marked as a motion sub-trajectory. Each motion sub-trajectory is expressed as a feature vector, the feature vector is input into the K-means algorithm for clustering analysis to determine the motion primitive, and based on each motion primitive, the predicted motion position of the tracking target at the current unmanned aerial vehicle autonomous tracking moment is determined.
[0013] Further, the determination process of the primary tracking target includes: If the motion mode of the tracking target is typical motion, the predicted motion position of the tracking target at the current unmanned aerial vehicle autonomous tracking moment is calculated respectively with the position distance deviation of multiple tracking similar targets, and the tracking similar target corresponding to the minimum position deviation is taken as the primary tracking target.
[0014] Further, the determination process of the primary tracking target further includes: If the movement mode of the tracking target is atypical movement, the predicted movement positions of the tracking target at the current autonomous tracking time of the UAV determined based on each movement primitive are respectively integrated with the probability values of the corresponding movement primitives to form a movement data group, wherein the probability value of the movement primitive is the frequency of occurrence of the movement primitive in the historical time; The actual positions of the tracking similar targets at the current time are obtained through images captured by the UAV in the current autonomous tracking, and position distance deviation calculations are respectively performed between the actual positions and the predicted movement positions contained in each movement data group, to obtain the position distance deviations corresponding to each movement data group; The probability values of the movement primitives contained in the movement data group are inverted and multiplied by the position distance deviations corresponding to the movement data group to obtain the position matching values of the movement data group; The position matching values of all movement data groups are averaged to obtain a tracking similarity value; The tracking similar targets are sorted according to the tracking similarity values from small to large, and the tracking similar target with the smallest tracking similarity value is selected as the primary tracking target.
[0015] Further, the determination of the tracking path is as follows: If the movement mode of the primary tracking target is typical movement, the predicted movement position of the primary tracking target at the next autonomous tracking time of the UAV is taken as the tracking terminal point, the actual position of the current UAV is taken as the tracking starting point, and the line connecting the tracking starting point and the tracking terminal point is the tracking path; If the movement mode of the primary tracking target is atypical movement, the probability values of the movement primitives corresponding to the predicted movement position of the primary tracking target at the next autonomous tracking time of the UAV are obtained, the predicted movement position corresponding to the maximum movement primitive probability value is selected as the target predicted movement position, the target predicted movement position is taken as the tracking terminal point, the actual position of the current UAV is taken as the tracking starting point, and the line connecting the tracking starting point and the tracking terminal point is the tracking path.
[0016] The beneficial effects of the present application are as follows: by tracking the feature similarity analysis of the target, it is determined whether multiple feature high-similarity tracking targets appear in the current unmanned aerial vehicle autonomous tracking; if so, by analyzing the running data of the tracking target in the historical multiple time, the motion mode of the tracking target is determined, the predicted motion position of the tracking target at the current unmanned aerial vehicle autonomous tracking time is determined according to the motion mode of the tracking target, and the position of multiple tracking similar targets is compared and analyzed to determine the primary tracking target, the predicted motion position of the primary tracking target at the next unmanned aerial vehicle autonomous tracking time is determined according to the motion mode of the primary tracking target, and the tracking path is determined according to the actual position of the current unmanned aerial vehicle. The present application solves the problem of tracking target confusion caused by adjacent single-frame tracking misjudgment, which leads to unmanned aerial vehicle tracking path planning error. Secondly, the present application also realizes the problem of tracking target motion mode identification. Through accurate identification of the motion mode, the identification accuracy of the tracking target and the accuracy of the tracking path planning are improved. BRIEF DESCRIPTION OF DRAWINGS
[0017] The present application will be further described below in conjunction with the drawings.
[0018] Figure 1 is a step flow chart of an unmanned aerial vehicle autonomous tracking trajectory planning method according to an embodiment of the present application; Figure 2 is a program block diagram of an unmanned aerial vehicle autonomous tracking trajectory planning system according to an embodiment of the present application. DETAILED DESCRIPTION
[0019] In order to make the technical means, creative features, purposes and effects realized by the present application easy to understand, the present application will be further described below in conjunction with specific embodiments.
[0020] The unmanned aerial vehicle autonomous tracking trajectory planning method according to the present application is mainly applied to the scene where multiple high-similarity tracking targets exist for the unmanned aerial vehicle, for example, when the unmanned aerial vehicle is performing autonomous tracking, the tracking target is normal in the last time, but in the next time, multiple tracking targets with high feature similarity exist, which leads to the unmanned aerial vehicle being unable to identify the target that needs to be tracked autonomously. Therefore, on the basis of solving this problem, the present application proposes an unmanned aerial vehicle autonomous tracking trajectory planning method.
[0021] Embodiment 1: Please refer to Figure 1 The unmanned aerial vehicle autonomous tracking trajectory planning method according to the present application embodiment includes the following steps: Step 1: According to the image captured by the current unmanned aerial vehicle autonomous tracking and the image captured by the historical previous time autonomous tracking, the feature similarity analysis of the tracking target is performed to determine whether multiple feature high-similarity tracking targets appear in the current unmanned aerial vehicle autonomous tracking; In step one, the feature similarity analysis process of the tracking target is as follows: The image feature extraction algorithm is used to extract feature points of the tracking target in the image taken by the unmanned aerial vehicle at the previous time point, and a feature descriptor is generated for each extracted feature point to form a reference feature descriptor set; The descriptor is a vector, which usually represents the gradient direction histogram of the pixels around the feature point or other features, and is used to describe the local image information of the feature point; Based on any target in the image taken by the unmanned aerial vehicle at the current time point; The image feature extraction algorithm is used to extract feature points of the target in the image taken by the unmanned aerial vehicle at the current time point, and a feature descriptor is generated for each extracted feature point to form a comparison feature descriptor set; The average similarity measure value between the reference feature descriptor set and the comparison feature descriptor set is calculated by using the Euclidean distance method; If the average similarity measure value is greater than or equal to the average similarity measure threshold, the tracking target is marked as a tracking similar target in the image taken by the unmanned aerial vehicle at the current time point; If the average similarity measure value is less than the average similarity measure threshold, no operation is performed; For example, the calculation process of the average similarity measure value between the reference feature descriptor set and the comparison feature descriptor set is exemplarily described: Suppose the reference feature descriptor set is {a1, a2}, and the comparison feature descriptor set is {b1, b2}; Wherein, a1=[1, 2], a2=[3, 4], b1=[2, 3], b2=[4, 5]; A plurality of descriptor pairs are obtained, which are s11=(a1, b1), s12=(a1, b2), s21=(a2, b1), and s22=(a2, b2); The Euclidean distance between each descriptor pair is calculated as follows: ; ; ; ;
[0022] The Euclidean distance between each descriptor pair is averaged to obtain the average similarity measure value, and the calculation method is as follows:
[0023] The number of similar targets is tracked. If the number of similar targets is 1, it indicates that the current unmanned aerial vehicle autonomous tracking does not appear multiple tracking targets with high similarity of features. If the number of similar targets is greater than 1, it indicates that the current unmanned aerial vehicle autonomous tracking appears multiple tracking targets with high similarity of features. Step two: if the current unmanned aerial vehicle autonomous tracking appears multiple tracking targets with high similarity of features, the running data of the tracking target in the historical multiple time is analyzed to determine the motion mode of the tracking target, wherein the motion mode includes typical motion and atypical motion. The running data includes the actual position, speed and acceleration data of the tracking target in the historical multiple time. In step two, the process of determining the motion mode of the tracking target is as follows: A typical motion model is established according to the kinematic equation, wherein the typical motion model includes but is not limited to: Uniform linear motion model: Wherein, (x0, y0) represents the initial position of the tracking target, (v x , v y ) represents the speed component of the tracking target, t is the time, and [x(t), y(t)] represents the position coordinates of the tracking target at time t. Uniform linear motion model: Wherein, (x0, y0) represents the initial position of the tracking target, (v 0x , v 0y ) represents the initial speed component of the tracking target, (a x , a y ) represents the acceleration component of the tracking target, t is the time, and [x(t), y(t)] represents the position coordinates of the tracking target at time t. Uniform circular motion model: Wherein, (x c , y c ) represents the center coordinates of the circular motion, R is the radius, ω is the angular velocity, φ is the initial phase, t is the time, and [x(t), y(t)] represents the position coordinates of the tracking target at time t. The actual position, speed and acceleration data of the tracking target in the historical multiple time are collected, and the least square method and other optimization algorithms are used to determine the parameters of the typical motion model. According to the determined parameters of the typical motion model, the predicted position of the tracking target output by the typical motion model at each historical time is calculated, and the absolute error between the predicted position and the actual position of the tracking target at each historical time is calculated to obtain the position error value. Exemplarily, the acquisition method of the position error value is exemplarily described: For example, assuming that the typical motion model is a uniform linear motion model, the actual position of the tracking target in the historical multiple time is: |History time t (unit s)|x coordinate|y coordinate| |0|0|0| |1|2|1| |2|4|2| |3|6|3| |4|8|4| The uniform linear motion model is where (x0, y0) represents the initial position of the tracking target, (v x , v y ) represents the velocity component of the tracking target, t is the time, and [x(t), y(t)] represents the position coordinates of the tracking target at time t; The least square method is used to estimate the model parameters (x0, y0, v x , v y ), that is, to minimize the sum of squares of errors between the position data and the model prediction value: where n is the number of historical time points, (x i , y i ) is the actual position of the tracking target at time t i ; The partial derivatives of S with respect to x0, y0, v x , v y are calculated, and set to 0 to obtain the following equation set: The solution is:
[0024] According to the solved model parameters, the predicted positions of the tracking target output by the typical motion model at each historical time are calculated, as follows: |History time t (unit s)|Predicted x coordinate|Predicted y coordinate| |0|0+2*0=0|0+1*0=0| |1|0+2*1=2|0+1*1=1| |2|0+2*2=4|0+1*2=2| |3|0+2*3=6|0+1*3=3| |4|0+2*4=8|0+1*4=4| The absolute error between the predicted position and the actual position of the tracking target at each historical time is calculated to obtain the position error value, for example: The position error value at history time 1 is: |2-2|+|1-1|=0; The position error value is compared with the preset position error range; If the position error value is within the preset position error range, the historical time corresponding to the position error value is marked as a typical motion time; If the position error value is not within the preset position error range, the historical moment corresponding to the position error value is marked as an atypical motion moment; The proportion of the duration of the atypical motion moment in the historical moment is counted to obtain an atypical motion maintenance value of the tracking target; Based on the atypical motion moment, the absolute difference proportion of the position error value at the atypical motion moment and the adjacent nearest preset position error range endpoint value is calculated to obtain an atypical motion degree value of the tracking target; For example, the absolute difference proportion of the position error value at the atypical motion moment and the adjacent nearest preset position error range endpoint value is calculated as follows: For example, the position error value at the atypical motion moment is ki, and the preset position error range is [kiy1, kiy2], wherein ki is not within the preset position error range [kiy1, kiy2], and ki is greater than kiy2. The absolute difference proportion is ; The atypical motion maintenance value and the atypical motion degree value are summed to obtain an atypical motion performance value; It should be noted that the physical meaning of the atypical motion performance value is that the atypical motion performance value is obtained by summing the atypical motion maintenance value and the atypical motion degree value, wherein the atypical motion maintenance value reflects the duration performance of the position error of the tracking target not within the preset position error range, the greater the atypical motion maintenance value, the higher the mismatch of the tracking target motion mode with the typical motion mode, and the greater the atypical motion degree value, also indicating that the tracking target motion mode does not match the typical motion mode; In some preferred embodiments, the atypical motion performance value is compared with an atypical motion performance threshold to determine whether the motion mode of the tracking target matches the typical motion model; If the atypical motion performance value is greater than or equal to the atypical motion performance threshold, it indicates that the motion mode of the tracking target does not match the typical motion model; If the atypical motion performance value is less than the atypical motion performance threshold, it indicates that the motion mode of the tracking target matches the typical motion model; According to the determination result of whether the motion mode of the tracking target matches the typical motion model, the motion mode of the tracking target is determined, specifically: If the motion mode of the tracking target matches any one of the typical motion models, the motion mode of the tracking target is a typical motion; If the motion mode of the tracking target does not match all the typical motion models, the motion mode of the tracking target is atypical motion; Step three: determining the predicted motion position of the tracking target at the current autonomous tracking time of the UAV according to the motion mode of the tracking target, and comparing and analyzing the positions of the multiple tracking similar targets, and determining the primary tracking target among the tracking similar targets; In step three, the process of determining the predicted motion position of the tracking target at the current autonomous tracking time of the UAV according to the motion mode of the tracking target is as follows: If the motion mode of the tracking target is typical motion, the current autonomous tracking time of the UAV is taken as input, and the predicted motion position of the tracking target is output through the typical motion model matched with the motion mode of the tracking target; For example, the motion mode of the tracking target is uniform linear motion, and the current tracking time of the UAV is the 30th second. The current tracking time (the 30th second) of the UAV is taken as input of the uniform linear motion model, and the two-dimensional coordinate position of the tracking target is output through the uniform linear motion model, that is, the predicted position of the tracking target at the tracking time of the 30th second of the UAV; If the motion mode of the tracking target is atypical motion, the predicted motion position at the current autonomous tracking time of the UAV is determined by using the motion primitive method, which specifically includes: According to the actual positions of the tracking target at multiple historical times, the motion trajectory of the tracking target at multiple historical times is drawn, and the motion trajectory of the tracking target at adjacent historical times is marked as a motion sub-trajectory. Each motion sub-trajectory is represented as a feature vector, which contains the speed and acceleration data of the tracking target; The feature vector is input into the K-means algorithm for clustering analysis. Each cluster represents a motion mode, i.e., a motion primitive; For example, assuming that there is a set of motion trajectory data of cars, each motion sub-trajectory contains speed and acceleration data, and the K-means algorithm is used for clustering. After clustering, three clusters are obtained: Cluster 1: small change in speed and acceleration close to 0, corresponding to the uniform linear motion primitive; Cluster 2: gradually increasing speed and positive acceleration, corresponding to the acceleration motion primitive; Cluster 3: gradually decreasing speed and negative acceleration, corresponding to the deceleration motion primitive; Based on each motion primitive, the predicted motion position of the tracking target at the current autonomous tracking time of the UAV is determined; It should be noted that if the movement of the tracking target is atypical, when determining the predicted movement position of the tracking target at the current UAV autonomous tracking moment based on the motion primitives, there are multiple predicted movement positions of the tracking target at the current UAV autonomous tracking moment, and each motion primitive corresponds to a predicted movement position; Exemplarily, the process of determining the predicted motion position of the tracking target at the current UAV autonomous tracking moment based on each motion primitive is as follows: Based on the uniform linear motion primitive, assuming that the position of the tracking target at the previous moment in history is (x T-1 ,y T-1 ), speed is v T-1 , where (T-1) is the previous moment in history, T is the current moment, and the direction is (the angle with the positive direction of the x-axis), the time interval is (the time interval between the previous moment and the current moment), then the predicted motion position of the current drone at the moment of autonomous tracking is:
[0025] Based on the accelerated motion primitive, assuming that the position of the tracking target at the previous moment in history is (x T-1 ,y T-1 ), speed is v T-1 , where (T-1) is the previous moment in history, T is the current moment, and the direction is (the angle with the positive direction of the x-axis), the time interval is , the acceleration is a T-1 , then the predicted motion position of the current UAV at the moment of autonomous tracking is:
[0026] In step 3, the process of determining the primary tracking target is as follows: If the movement of the tracking target is typical, the position deviations between the predicted movement position of the tracking target at the current UAV autonomous tracking moment and the positions of multiple tracking similar targets are calculated, and the tracking similar target corresponding to the minimum position deviation is taken as the primary tracking target; For example, the position distance deviation is calculated as follows: Assuming that the predicted motion position of the tracking target is (x1, y1), and there is a tracking target with high feature similarity at (x2, y2), the calculation formula for the position distance deviation is: If the target's movement is atypical, the primary target is determined as follows: The predicted motion position of the tracking target determined based on each motion primitive at the current autonomous tracking time of the unmanned aerial vehicle is integrated with the probability value of the corresponding motion primitive to form a motion data group, wherein each motion data group includes a predicted motion position and a probability value of a corresponding motion primitive; wherein the probability value of the motion primitive is the frequency of occurrence of the motion primitive within a historical time period; Based on any one tracking similar target, the actual position of the tracking similar target at the current time is obtained through the image captured by the current autonomous tracking of the unmanned aerial vehicle, and the position distance deviation calculation is performed with the predicted motion position included in each motion data group to obtain the position distance deviation corresponding to each motion data group; Based on any one motion data group, the reciprocal of the probability value of the motion primitive included in the motion data group is calculated, and the product operation is performed with the position distance deviation corresponding to the motion data group to obtain the position matching value of the motion data group; The position matching values of all motion data groups are averaged to obtain a tracking similarity value, wherein the tracking similarity value reflects the degree of position deviation between the tracking similar target and the tracking target; The tracking similar targets are sorted according to the tracking similarity value from small to large, and the tracking similar target with the smallest tracking similarity value is selected as the primary tracking target; It should be noted that the physical meaning represented by the tracking similarity value is that the tracking similarity value is obtained by averaging the position matching values of all motion data groups, and the position matching value is obtained by calculating the reciprocal of the probability value of the motion primitive included in the motion data group and the product operation with the position distance deviation corresponding to the motion data group. It should be noted that the position distance deviation corresponding to the motion data group reflects the distance deviation between the predicted motion position of the tracking target and the actual position of the tracking similar target at the current autonomous tracking time of the unmanned aerial vehicle. The smaller the distance deviation, the more matched the position of the tracking similar target and the tracking target, and the more likely the tracking similar target is the tracking target. The significance of calculating the reciprocal of the probability value of the motion primitive included in the motion data group and the product operation with the position distance deviation corresponding to the motion data group is that the probability value of the motion primitive included in the motion data group reflects the probability of the predicted motion position of the tracking target appearing in the motion data group. After taking the reciprocal, the physical relationship corresponding to the position distance deviation is consistent, i.e. the smaller the position distance deviation, the smaller the probability value of the motion primitive after taking the reciprocal, which means that the position of the tracking similar target is closer to the predicted position of the tracking target. Therefore, the smaller the tracking similarity value, the closer the actual position of the tracking similar target to the predicted position of the tracking target. Step four: determining the predicted motion position of the primary tracking target at the next autonomous tracking time of the unmanned aerial vehicle according to the motion mode of the primary tracking target, and determining the tracking path in combination with the actual position of the current unmanned aerial vehicle; In step four, the predicted motion position of the primary tracking target at the next unmanned aerial vehicle autonomous tracking time is determined according to the motion mode of the primary tracking target, and the predicted motion position of the tracking target at the current unmanned aerial vehicle autonomous tracking time is determined according to the motion mode of the tracking target. The process of determining the tracking path in combination with the actual position of the current unmanned aerial vehicle is as follows: In step four, the process of determining the tracking path in combination with the actual position of the current unmanned aerial vehicle is as follows: If the motion mode of the primary tracking target is typical motion, the predicted motion position of the primary tracking target at the next unmanned aerial vehicle autonomous tracking time is taken as the tracking end point, the actual position of the current unmanned aerial vehicle is taken as the tracking start point, and the line connecting the tracking start point and the tracking end point is the tracking path. If the motion mode of the primary tracking target is non-typical motion, the probability values of the motion primitives corresponding to the predicted motion position of the primary tracking target at the next unmanned aerial vehicle autonomous tracking time are obtained, the predicted motion position corresponding to the maximum motion primitive probability value is taken as the target predicted motion position, the target predicted motion position is taken as the tracking end point, the actual position of the current unmanned aerial vehicle is taken as the tracking start point, and the line connecting the tracking start point and the tracking end point is the tracking path.
[0027] The technical scheme of the embodiment of the present application is as follows: the feature similarity of the tracking target is analyzed to determine whether multiple tracking targets with high feature similarity appear in the current unmanned aerial vehicle autonomous tracking; if so, the motion mode of the tracking target is determined through analysis of the running data of the tracking target at multiple historical moments, the predicted motion position of the tracking target at the current unmanned aerial vehicle autonomous tracking time is determined according to the motion mode of the tracking target, the position of the multiple tracking similar targets is compared and analyzed to determine the primary tracking target, the predicted motion position of the primary tracking target at the next unmanned aerial vehicle autonomous tracking time is determined according to the motion mode of the primary tracking target, and the tracking path is determined according to the actual position of the current unmanned aerial vehicle. The present application solves the problem of tracking target confusion caused by misjudgment of adjacent single-frame tracking, which leads to errors in unmanned aerial vehicle tracking path planning. In addition, the present application also realizes the identification of the motion mode of the tracking target. Through accurate identification of the motion mode, the identification accuracy of the tracking target and the accuracy of the tracking path planning are improved. Specifically, step one determines whether multiple tracking targets with high similarity appear by comparing the feature information of the target in the current image with the feature information of the target in the historical image in real time, which effectively avoids tracking target confusion caused by misjudgment. Step two divides the motion mode of the tracking target into two categories, typical motion and non-typical motion, by analyzing the motion trajectory characteristics of the target in the time dimension, which improves the identification ability of complex motion targets and is suitable for target differentiation in dynamic scenes. Step three ensures the scientificity and robustness of the selection of the primary tracking target by matching the position of the tracking target in combination with the determined motion mode, which effectively filters the mis-tracking caused by temporary similarity. Step four realizes the planning of the tracking path after the primary tracking target is determined.
[0028] Embodiment 2: please refer to Figure 2 As shown in the figure, the unmanned aerial vehicle autonomous tracking trajectory planning system according to the embodiment of the application comprises the following modules: The tracking target similarity analysis module: according to the image captured by the current autonomous tracking of the unmanned aerial vehicle and the image captured by the previous autonomous tracking of the unmanned aerial vehicle, the feature similarity of the tracking target is analyzed to determine whether multiple tracking targets with high feature similarity appear in the current autonomous tracking of the unmanned aerial vehicle; The process of feature similarity analysis of the tracking target is as follows: The image feature extraction algorithm is used to extract feature points of the tracking target in the image captured by the previous autonomous tracking of the unmanned aerial vehicle, and a feature descriptor is generated for each extracted feature point to form a reference feature descriptor set; Wherein, the descriptor is a vector, which usually represents the gradient direction histogram of the pixels around the feature point or other features, used to describe the local image information of the feature point; Based on any target in the image captured by the current autonomous tracking of the unmanned aerial vehicle; The image feature extraction algorithm is used to extract feature points of the target in the image captured by the current autonomous tracking of the unmanned aerial vehicle, and a feature descriptor is generated for each extracted feature point to form a comparison feature descriptor set; The average similarity measure value between the reference feature descriptor set and the comparison feature descriptor set is calculated by using the Euclidean distance method; If the average similarity measure value is greater than or equal to the average similarity measure threshold, the target in the image captured by the current autonomous tracking of the unmanned aerial vehicle is marked as a tracking similar target; If the average similarity measure value is less than the average similarity measure threshold, no operation is performed; The number of tracking similar targets is counted, if the number of tracking similar targets is 1, it indicates that multiple tracking targets with high feature similarity do not appear in the current autonomous tracking of the unmanned aerial vehicle, if the number of tracking similar targets is greater than 1, it indicates that multiple tracking targets with high feature similarity appear in the current autonomous tracking of the unmanned aerial vehicle; The motion mode analysis module: if multiple tracking targets with high feature similarity appear in the current autonomous tracking of the unmanned aerial vehicle, the motion mode of the tracking target is determined through the running data analysis of the tracking target in the historical multiple time, wherein the motion mode includes typical motion and atypical motion; Wherein, the running data includes the actual position, speed and acceleration data of the tracking target in the historical multiple time; The process of determining the motion mode of the tracking target is as follows: The typical motion model is established according to the kinematic equation, wherein the typical motion model includes but is not limited to: Uniform linear motion model: wherein (x0, y0) represents an initial position of the tracking target, (v x , v y ) represents a velocity component of the tracking target, t is time, and [x(t), y(t)] represents a position coordinate of the tracking target at time t; uniformly accelerated linear motion model: wherein (x0, y0) represents an initial position of the tracking target, (v 0x , v 0y ) represents an initial velocity component of the tracking target, (a x , a y ) represents an acceleration component of the tracking target, t is time, and [x(t), y(t)] represents a position coordinate of the tracking target at time t; uniformly accelerated linear motion model: wherein (x c , y c ) represents a center coordinate of a circle of circular motion, R is a radius, ω is an angular velocity, φ is an initial phase, t is time, and [x(t), y(t)] represents a position coordinate of the tracking target at time t; Using the actual position, velocity, and acceleration data of the tracking target collected at multiple historical moments, a typical motion model parameter is determined using an optimization algorithm such as the least squares method, and a predicted position of the tracking target output by the typical motion model at each historical moment is calculated based on the determined typical motion model parameter, an absolute error between the predicted position and the actual position of the tracking target at each historical moment is calculated, and a position error value is obtained; The position error value is compared with a preset position error range; If the position error value is within the preset position error range, the historical moment corresponding to the position error value is marked as a typical motion moment; If the position error value is not within the preset position error range, the historical moment corresponding to the position error value is marked as an atypical motion moment; The proportion of the duration of the atypical motion moment in the historical moments is calculated, and an atypical motion maintenance value of the tracking target is obtained; Based on the atypical motion moment, an absolute difference ratio of the position error value at the atypical motion moment and the endpoint value of the adjacent nearest preset position error range is calculated, an atypical motion error out-of-range ratio is obtained, and the atypical motion error out-of-range ratio of all atypical motion moments is averaged to obtain an atypical motion degree value of the tracking target; The atypical motion maintenance value and the atypical motion degree value are summed to obtain an atypical motion performance value; In some preferred embodiments, the atypical motion performance value is compared with an atypical motion performance threshold value to determine whether the motion mode of the tracking target matches the typical motion model; if the atypical motion performance value is greater than or equal to the atypical motion performance threshold value, it indicates that the motion mode of the tracking target does not match the typical motion model; if the atypical motion performance value is less than the atypical motion performance threshold value, it indicates that the motion mode of the tracking target matches the typical motion model; According to the judgment result of whether the motion mode of the tracking target matches the typical motion model, the motion mode of the tracking target is determined, specifically: if the motion mode of the tracking target matches any one of the typical motion models, the motion mode of the tracking target is typical motion; if the motion mode of the tracking target does not match all the typical motion models, the motion mode of the tracking target is atypical motion; The target tracking determination module determines the primary tracking target among the tracking similar targets according to the predicted motion position of the tracking target at the current unmanned aerial vehicle autonomous tracking time. The process of determining the predicted motion position of the tracking target at the current unmanned aerial vehicle autonomous tracking time according to the motion mode of the tracking target is: if the motion mode of the tracking target is typical motion, the current unmanned aerial vehicle autonomous tracking time is taken as input, and the predicted motion position of the tracking target is output through the typical motion model matched by the motion mode of the tracking target; if the motion mode of the tracking target is atypical motion, the predicted motion position of the tracking target at the current unmanned aerial vehicle autonomous tracking time is determined by using the motion primitive method, specifically including: According to the actual positions of the tracking target at the historical multiple times, the motion trajectory of the tracking target at the multiple historical times is drawn, and the motion trajectory of the tracking target at adjacent historical times is marked as a motion sub-trajectory. Each motion sub-trajectory is represented as a feature vector, which contains the speed and acceleration data of the tracking target; The feature vector is input into the K-means algorithm for clustering analysis. Each cluster represents a motion mode, i.e., a motion primitive; Based on each motion primitive, the predicted motion position of the tracking target at the current unmanned aerial vehicle autonomous tracking time is determined; In step three, the determination process of the primary tracking target is: if the motion mode of the tracking target is typical motion, the predicted motion position of the tracking target at the current unmanned aerial vehicle autonomous tracking time is calculated respectively, and the position distance deviation of the multiple tracking similar targets is calculated. The tracking similar target corresponding to the minimum position deviation is taken as the primary tracking target; if the motion mode of the tracking target is atypical motion, the determination method of the primary tracking target is: The predicted motion position of the tracking target determined based on each motion primitive at the current autonomous tracking time of the UAV and the probability value of the corresponding motion primitive are integrated into a motion data group, wherein each motion data group includes a predicted motion position and a probability value of a corresponding motion primitive; wherein the probability value of the motion primitive is the frequency of occurrence of the motion primitive within a historical time period; Based on any one tracking similar target, the actual position of the tracking similar target at the current time is obtained through the image captured by the current autonomous tracking of the UAV, and the position distance deviation calculation is performed between the actual position and the predicted motion position included in each motion data group, to obtain the position distance deviation corresponding to each motion data group; Based on any one motion data group, the reciprocal of the probability value of the motion primitive included in the motion data group is obtained, and the reciprocal is multiplied by the position distance deviation corresponding to the motion data group to obtain the position matching value of the motion data group; The position matching values of all motion data groups are averaged to obtain a tracking similarity value, wherein the tracking similarity value reflects the degree of position deviation between the tracking similar target and the tracking target; The tracking similar targets are sorted according to the tracking similarity value from small to large, and the tracking similar target with the smallest tracking similarity value is selected as the primary tracking target; The tracking path generation module: determines the predicted motion position of the primary tracking target at the next autonomous tracking time of the UAV according to the motion mode of the primary tracking target, and determines the tracking path according to the actual position of the current UAV; The way of determining the predicted motion position of the primary tracking target at the next autonomous tracking time of the UAV according to the motion mode of the primary tracking target is the same as the way of determining the predicted motion position of the tracking target at the current autonomous tracking time of the UAV according to the motion mode of the tracking target, which will not be described here; The process of determining the tracking path according to the actual position of the current UAV is as follows: If the motion mode of the primary tracking target is typical motion, the predicted motion position of the primary tracking target at the next autonomous tracking time of the UAV is taken as the tracking endpoint, the actual position of the current UAV is taken as the tracking starting point, and the line connecting the tracking starting point and the tracking endpoint is the tracking path; If the motion mode of the primary tracking target is non-typical motion, the probability value of the motion primitive corresponding to the predicted motion position of the primary tracking target at the next autonomous tracking time of the UAV is obtained, the predicted motion position corresponding to the maximum motion primitive probability value is selected as the target predicted motion position, the target predicted motion position is taken as the tracking endpoint, the actual position of the current UAV is taken as the tracking starting point, and the line connecting the tracking starting point and the tracking endpoint is the tracking path.
[0029] The above shows and describes the basic principles, main features and advantages of the present application. Those skilled in the art should understand that the present application is not limited to the above-mentioned embodiments, and the above-mentioned embodiments and descriptions in the specification are only to illustrate the principles of the present application. Without departing from the spirit and scope of the present application, various changes and improvements can be made to the present application, and these changes and improvements all fall within the scope of the claimed present application. The scope of protection of the present application is defined by the appended claims and their equivalents.
Claims
1. A method for autonomous tracking trajectory planning of a UAV, characterized by: The following steps are involved: Based on the images taken by the drone during the current autonomous tracking and the images taken during the previous autonomous tracking, the feature similarity analysis of the tracking target is performed to determine whether there are multiple tracking targets with high feature similarity in the current autonomous tracking of the drone. If it appears, the movement mode of the tracking target is determined by analyzing the running data of the tracking target at multiple historical moments, where the movement mode includes typical movement and atypical movement; Determine the predicted movement position of the target at the current autonomous tracking moment based on the target's movement pattern, and compare and analyze it with the positions of multiple similar targets. Determine the primary target among the similar targets. The predicted motion position of the primary tracking target at the next UAV autonomous tracking moment is determined according to the movement mode of the primary tracking target, and the tracking path is determined in combination with the actual position of the current UAV.
2. The method for autonomous tracking trajectory planning of a UAV according to claim 1, characterized in that: The process of analyzing the similarity of the features of the tracking target is as follows: The image feature extraction algorithm is used to perform feature processing on the images taken by the current autonomous tracking of the UAV and the images taken by the autonomous tracking at the previous moment in history, and a comparison feature descriptor set and a reference feature descriptor set are obtained; The average similarity measure between the reference feature descriptor set and the comparison feature descriptor set is calculated using the Euclidean distance method. If the average similarity measure is greater than or equal to the average similarity measure threshold, the tracking target is marked as a tracking similar target in the image currently taken by the UAV autonomous tracking; If the number of similar tracking targets is greater than 1, it means that the current UAV is autonomously tracking multiple targets with high similarity in features.
3. The method for autonomous tracking trajectory planning of a UAV according to claim 2, characterized in that: The comparison feature descriptor set and the reference feature descriptor set are obtained in the following manner: An image feature extraction algorithm is used to extract feature points of the tracking target in the image taken by the UAV during autonomous tracking at the previous moment in history, and a feature descriptor is generated for each extracted feature point to form a reference feature descriptor set; An image feature extraction algorithm is used to extract feature points of the target in the image currently captured by the UAV for autonomous tracking, and a feature descriptor is generated for each extracted feature point to form a comparison feature descriptor set.
4. The method for autonomous tracking trajectory planning of a UAV according to claim 1, characterized in that: The process of determining the movement mode of the tracking target is specifically as follows: According to the typical motion model established by the kinematic equation, the absolute error between the predicted position and the actual position of the tracking target at each historical moment is calculated to obtain the position error value; If the position error value is not within the preset position error range, then determining the atypical motion moment of the tracking target; Through the processing and analysis of atypical movement moments, the atypical movement maintenance value and the atypical movement degree value are obtained, and the sum of the values is used to obtain the atypical movement performance value; The atypical motion performance value is used to determine whether the motion pattern of the tracked target matches the typical motion model; If the motion mode of the tracking target matches any typical motion model, the motion mode of the tracking target is a typical motion; If the movement mode of the tracking target does not match all typical movement models, the movement mode of the tracking target is atypical.
5. The method for autonomous tracking trajectory planning of a UAV according to claim 4, characterized in that: The specific method of obtaining the position error value is: The actual position, velocity and acceleration data of the tracking target collected at multiple historical moments are used to determine the parameters of the typical motion model using optimization algorithms such as the least squares method. The predicted position of the tracking target output by the typical motion model at each historical moment is calculated, and the absolute error between the predicted position and the actual position of the tracking target at each historical moment is calculated to obtain the position error value.
6. The method for autonomous tracking trajectory planning of a UAV according to claim 4, characterized in that: The atypical exercise maintenance value is obtained as follows: Count the proportion of atypical movement moments in historical moments to obtain the atypical movement maintenance value of the tracked target; The atypical exercise level value is obtained as follows: Based on the atypical operating moments, the absolute difference ratio between the position error value at the atypical operating moment and the nearest endpoint value of the preset position error range is calculated to obtain the atypical operating moment position error exceeding limit ratio. The position error exceeding limit ratios of all atypical operating moments are averaged to obtain the atypical motion degree value of the tracking target.
7. The method for autonomous tracking trajectory planning of a UAV according to claim 1, characterized in that: The process of determining the predicted motion position of the tracking target at the current UAV autonomous tracking moment according to the motion mode of the tracking target is as follows: If the target's motion is typical, the current autonomous tracking moment is used as input, and the predicted motion position of the target is output through the typical motion model that matches the target's motion. If the movement of the tracking target is atypical, the movement trajectory of the tracking target in multiple historical moments is drawn based on the actual position of the tracking target in multiple historical moments, and the movement trajectory of the tracking target in adjacent historical moments is marked as a movement sub-trajectory. Each movement sub-trajectory is represented as a feature vector. The feature vector is input into the K-means algorithm for cluster analysis to determine the motion primitives. Based on each motion primitive, the predicted movement position of the tracking target at the current UAV autonomous tracking moment is determined.
8. The method for autonomous tracking trajectory planning of a UAV according to claim 1, characterized in that: The process of determining the primary tracking target includes: If the movement mode of the tracking target is typical, the position distance deviations between the predicted movement position of the tracking target at the current UAV autonomous tracking moment and multiple tracking similar targets are calculated, and the tracking similar target corresponding to the minimum position deviation is taken as the primary tracking target.
9. The method for autonomous tracking trajectory planning of a UAV according to claim 8, characterized in that: The process of determining the primary tracking target also includes: If the movement of the target is atypical, the predicted movement position of the target at the current autonomous tracking moment determined by each motion primitive is combined with the probability value of the corresponding motion primitive to form a motion data set, where the probability value of the motion primitive is the frequency of occurrence of the motion primitive in the historical moment. The actual position of the target at the current moment is obtained through the image captured by the current autonomous tracking of the UAV, and the position distance deviation is calculated with the predicted motion position contained in each motion data group to obtain the position distance deviation corresponding to each motion data group; The probability value of the motion primitive contained in the motion data group is calculated by taking the inverse thereof and then multiplying it with the position distance deviation corresponding to the motion data group to obtain the position matching value of the motion data group; The position matching values of all motion data groups are averaged to obtain the tracking similarity value; Sort the tracking similar targets in ascending order according to the tracking similarity value, and select the tracking similar target with the first ranking as the primary tracking target.
10. The method for autonomous tracking trajectory planning of a UAV according to claim 1, characterized in that: The tracking path is determined as follows: If the primary tracking target's movement is typical, the predicted movement position of the primary tracking target at the next UAV autonomous tracking moment is used as the tracking end point, and the current UAV's actual position is used as the tracking start point. The line connecting the tracking start point and the tracking end point is the tracking path. If the movement mode of the primary tracking target is atypical, the probability values of the motion primitives corresponding to the predicted motion position of the primary tracking target at the next UAV autonomous tracking moment are obtained respectively, and the predicted motion position corresponding to the largest motion primitive probability value is selected as the target predicted motion position. The target predicted motion position is used as the tracking end point, and the actual position of the current UAV is used as the tracking starting point. The line connecting the tracking starting point and the tracking end point is the tracking path.