A method for determining the correlation of UAV trajectories in monitoring blind areas

Abstract

The present invention discloses a method for determining the correlation of unmanned aerial vehicle trajectories for monitoring blind spots. The method comprises: obtaining the monitoring blind spot boundary of the current unmanned aerial vehicle monitoring space after data processing of the historical flight data of the unmanned aerial vehicle monitored and obtained in the unmanned aerial vehicle monitoring space; monitoring and obtaining each flight trajectory of the unmanned aerial vehicle in a short period of time; and then judging whether the endpoints of each flight trajectory of the unmanned aerial vehicle in a short period of time are at the monitoring blind spot boundary; if the endpoints of the two adjacent flight trajectories in the short period of time are both at the monitoring blind spot boundary, then predicting the trajectory of the current two flight trajectories and then judging the correlation of the predicted trajectories of the current two flight trajectories; if the predicted trajectories are correlated, then connecting the trajectories of the current two flight trajectories; and finally using the unmanned aerial vehicle control area as the starting point and end point area of ​​the unmanned aerial vehicle to obtain the complete flight trajectory of the unmanned aerial vehicle. The present invention effectively identifies whether the unmanned aerial vehicle passing through the monitoring blind spot is the same unmanned aerial vehicle.

Description

Technical Field

[0001] The present invention relates to a method for processing unmanned aerial vehicle (UAV) trajectories in the field of unmanned aerial vehicle (UAV) supervision, and in particular to a method for determining the relevance of UAV trajectories in a monitoring blind area. Background Art

[0002] At present, the field of drone control is becoming more and more mature, but the field of drone supervision is relatively backward. In drone supervision, cameras are usually used to identify and monitor drones, but when drone targets are small or similar, it is difficult for cameras to distinguish drones. When calculating the number of drones flying in a space, a large error will occur. In order to better monitor the number of drones, a method based on trajectory correlation judgment must be adopted to determine whether the drones flying through the monitoring blind spot are the same. Summary of the invention

[0003] In order to solve the problems and needs in the background technology, the purpose of the present invention is to propose a method for determining the correlation of UAV trajectories for monitoring blind spots. The present invention first uses the long-term historical flight data of the UAV to determine the location and boundary of the monitoring blind spot, and then performs forward and backward trajectory prediction on the trajectory segments near the boundary of the monitoring blind spot, and then determines whether the predicted trajectories are related. If they are related, it means that the flight trajectory segments near the boundary of the monitoring blind spot are from the same UAV.

[0004] The technical solution of the present invention is:

[0005] The present invention comprises the following steps:

[0006] S1: According to the preset time interval, the historical flight data of the UAV obtained by monitoring in the UAV monitoring space is segmented to obtain the corresponding historical flight data segments, then the endpoints of each historical flight data segment are taken to form a historical flight data endpoint set, and then the historical flight data endpoint set is clustered to fit the monitoring blind area boundary of the current UAV monitoring space;

[0007] S2: In the current drone monitoring space, monitor and obtain the drone flight data in the short term, segment the drone flight data in the short term at preset time intervals to obtain the corresponding short-term flight data segments, and then perform trajectory fitting on each short-term flight data segment to obtain each flight trajectory of the drone in the short term;

[0008] S3: Take the endpoints of each segment of the UAV's flight trajectory in a short period of time and form a short-term flight data endpoint set, cluster the short-term flight data endpoint set, cluster to obtain the UAV control area, and at the same time, judge whether the endpoints of each segment of the UAV's flight trajectory in a short period of time are within the monitoring blind zone boundary according to the monitoring blind zone boundary of the current UAV monitoring space;

[0009] S4: For two adjacent flight trajectories in a short period of time, if the endpoints of the two adjacent flight trajectories in a short period of time are both at the monitoring blind zone boundary, the trajectory of the current two adjacent flight trajectories is predicted to obtain the corresponding predicted trajectory, and then the correlation of the predicted trajectories of the current two adjacent flight trajectories is determined; if the predicted trajectories are related, it is considered that the current two adjacent flight trajectories are from the same UAV; then the trajectories of the current two adjacent flight trajectories are connected according to the predicted trajectories of the current two adjacent flight trajectories; if the predicted trajectories are not related, it is considered that they are not from the same UAV; if the endpoints of the two adjacent flight trajectories in a short period of time are not at the monitoring blind zone boundary, no processing is performed;

[0010] S5: Repeat S4 to judge the two adjacent flight trajectories remaining in the short term, and obtain the continuous flight trajectory of the drone in the current drone monitoring space in the short term; then use the corresponding drone control area as the starting point-end point area of ​​the drone to obtain the complete flight trajectory of the drone in the current drone monitoring space.

[0011] In the S1, the endpoint set of historical flight data is clustered to obtain multiple historical clusters, the x-axis direction is set to the horizontal direction, the center point coordinates of the multiple historical clusters are projected to the x-axis, the two projection points x0 and x1 with the farthest distance on the x-axis are selected and recorded as the first and second projection points, the line segment connecting the center point coordinates of the historical clusters corresponding to the first and second projection points is used as the diameter and the first sphere S0 is made with the center of the current line segment as the sphere center, and the outer spherical surface of the first sphere S0 is used as the maximum boundary of the monitoring blind area;

[0012] Next, a straight line l is made that passes through the center of the first sphere S0 and is perpendicular to the x-axis. Two projection points x2 and x3 that are closest to the straight line l are found on the x-axis on both sides of the straight line l and are recorded as the third and fourth projection points. The line segment connecting the coordinates of the center points of the historical clusters corresponding to the third and fourth projection points is used as the diameter and a second sphere S1 is made with the center of the current line segment as the center. The outer surface of the second sphere S1 is used as the minimum boundary of the monitoring blind spot; the annular area between the maximum boundary and the minimum boundary of the monitoring blind spot is used as the boundary of the monitoring blind spot.

[0013] The S3 is specifically:

[0014] S31: Filter each segment of the UAV's flight trajectory in the short term through the Kalman inverse filter to obtain a discrete flight data sequence corresponding to each segment of the flight trajectory, take the endpoint data of each discrete flight data sequence and form a short-term flight data endpoint set;

[0015] S32: Cluster the endpoint set of short-term flight data to obtain multiple short-term clusters, and use the space where the short-term clusters formed by clustering the endpoint data at the take-off point and landing point of the drone are located as the drone control area, and determine whether the center points of the remaining short-term clusters except the drone control area are in the monitoring blind area boundary; if the center point of each short-term cluster is in the monitoring blind area boundary, then the endpoint data of the current short-term cluster is clustered in the monitoring blind area boundary, and the endpoint data of the current short-term cluster is in the monitoring blind area boundary, which means that the endpoints of the corresponding flight trajectories are in the monitoring blind area boundary; otherwise, it is not in the monitoring blind area boundary.

[0016] In S4, the Hausdorff distance method is used to calculate the Hausdorff distance between the predicted trajectories of the current two preceding and following flight trajectories. If the Hausdorff distance between the predicted trajectories of the current two preceding and following flight trajectories is less than the preset distance, the predicted trajectories of the current two preceding and following flight trajectories are correlated; otherwise, they are not correlated.

[0017] The preset time intervals in S1 and S2 are the same.

[0018] In S2, a Kalman filter is used to perform trajectory fitting on each short-term flight data segment.

[0019] In S4, the predicted trajectories of the two current flight trajectories are averaged to obtain a final predicted trajectory, and the two current flight trajectories are connected to the final predicted trajectory. The beneficial effects of the present invention are:

[0020] The present invention can effectively determine the monitoring blind area and the blind area boundary in the space, determine whether multiple trajectory segments near the monitoring blind area come from the same drone, and thus accurately calculate the number of drones in the space, and has broad application prospects in the field of drone supervision. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 This is a flowchart for determining the relevance of the UAV flight trajectory;

[0022] Figure 2 This is a schematic diagram of fitting the blind area boundary;

[0023] Figure 3 Schematic diagram of drone trajectory merging. DETAILED DESCRIPTION

[0024] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments:

[0025] like Figure 1 As shown, the present invention comprises the following steps:

[0026] S1: According to the preset time interval, the historical flight data of multiple drones obtained by long-term monitoring in the drone monitoring space are segmented to obtain the corresponding historical flight data segments. Then, the endpoints of each historical flight data segment are taken to form a historical flight data endpoint set. The endpoints of each historical flight data segment are composed of multiple data points at each end of the historical flight data segment. Then, the historical flight data endpoint set is clustered to fit the monitoring blind area boundary of the current drone monitoring space, such as Figure 2 As shown; the endpoints are the starting point and the end point of the flight trajectory corresponding to the current historical flight data segment and do not include the take-off point and the landing point of the UAV;

[0027] In S1, the endpoint set of historical flight data is clustered to obtain multiple historical clustering clusters. In the specific implementation, the number of historical clustering clusters is 2n, and the center point of the cluster represents the starting point or end point of the flight data, where n is the number of historical flight data segments. The x-axis direction is set to the horizontal direction, and the center point coordinates of multiple historical clustering clusters are projected to the x-axis. The two projection points x0 and x1 with the farthest distance on the x-axis are selected and recorded as the first and second projection points. The line segment connecting the center point coordinates of the historical clustering clusters corresponding to the first and second projection points is used as the diameter, and the first sphere S0 is made with the center of the current line segment as the center of the sphere. The outer spherical surface of the first sphere S0 is used as the maximum boundary of the monitoring blind area; the drone monitoring space outside the first sphere S0 is the monitoring visible area.

[0028] Next, a straight line l is made that passes through the center of the first sphere S0 and is perpendicular to the x-axis. Two projection points x2 and x3 that are closest to the straight line l are found on the x-axis on both sides of the straight line l and are recorded as the third and fourth projection points. The line segment connecting the coordinates of the center points of the historical clusters corresponding to the third and fourth projection points is used as the diameter and a second sphere S1 is made with the center of the current line segment as the center. The outer surface of the second sphere S1 is used as the minimum boundary of the monitoring blind spot. The drone monitoring space within the second sphere S1 is the monitoring blind spot. The annular area between the maximum boundary and the minimum boundary of the monitoring blind spot is used as the boundary of the monitoring blind spot.

[0029] S2: In the current UAV monitoring space, monitor and obtain the UAV flight data in the short term, segment the UAV flight data in the short term at preset time intervals, obtain the corresponding short-term flight data segments, and then perform trajectory fitting for each short-term flight data segment in chronological order to obtain the flight trajectories of each segment of the UAV in the short term; in the specific implementation, the short term refers to the UAV flying for 2 hours in the current UAV monitoring space.

[0030] The preset time intervals in S1 and S2 are the same. In a specific implementation, the preset time interval is t d , the time interval between adjacent data points in the historical flight data is t ’ d , satisfying kt’ d ≤t d , k≥10, k is the time coefficient. Segment trajectory id n Starting point end n=1, 2, 3..., m≥1.

[0031] In S2, a Kalman filter is used to fit the trajectory of each short-term flight data segment.

[0032] S3: Take the endpoints of each segment of the UAV's flight trajectory in a short period of time and form a short-term flight data endpoint set, cluster the short-term flight data endpoint set, cluster to obtain the UAV control area, the UAV control area is at least one, and at the same time, judge whether the endpoints of each segment of the UAV's flight trajectory in a short period of time are at the monitoring blind area boundary according to the monitoring blind area boundary of the current UAV monitoring space;

[0033] S3 is specifically:

[0034] S31: Filter each segment of the UAV's flight trajectory in a short period of time through an inverse Kalman filter to obtain a discrete flight data sequence corresponding to each segment of the flight trajectory, take the endpoint data of each discrete flight data sequence and form a short-term flight data endpoint set; the endpoint data of each discrete flight data sequence is composed of multiple data points at the starting end and multiple data points at the ending end.

[0035] S32: Cluster the endpoint set of short-term flight data to obtain multiple short-term clusters. In the specific implementation, the number of short-term clusters is 2N, where N is the number of short-term flight data segments. The space where the short-term clusters formed by clustering the endpoint data at the take-off point and landing point of the drone are located is used as the drone control area S. p , it is determined that except for the drone control area S p Whether the center points of the remaining short-term clusters other than the short-term clusters are in the monitoring blind zone boundary; if the center point of each short-term cluster is in the monitoring blind zone boundary, the endpoint data of the clustered current short-term clusters are all in the monitoring blind zone boundary, and the endpoint data of the current short-term clusters are all in the monitoring blind zone boundary, which means that the endpoints of the corresponding flight trajectories are in the monitoring blind zone boundary; otherwise, they are not in the monitoring blind zone boundary;

[0036] S4: Figure 3As shown, for two adjacent flight trajectories in a short period of time, if the endpoints of the two adjacent flight trajectories in a short period of time are both at the monitoring blind zone boundary, specifically: the end point of the current front flight trajectory and the starting point of the rear flight trajectory are both at the monitoring blind zone boundary, then the trajectory of the current two adjacent flight trajectories is predicted to obtain the corresponding predicted trajectory, the predicted trajectory is within the monitoring blind zone boundary, and then the correlation of the predicted trajectories of the current two adjacent flight trajectories is determined; if the predicted trajectories are related, it is considered that the current two adjacent flight trajectories are from the same UAV; then the current two adjacent flight trajectories are connected according to the predicted trajectories of the current two adjacent flight trajectories; if the predicted trajectories are not related, it is considered that they are not from the same UAV, and then no processing is performed; if the endpoints of the two adjacent flight trajectories in a short period of time are not at the monitoring blind zone boundary, no processing is performed;

[0037] In S4, the Hausdorff distance method is used to calculate the Hausdorff distance between the predicted trajectories of the current two flight trajectories. If the Hausdorff distance between the predicted trajectories of the current two flight trajectories is less than a preset distance, in a specific implementation, the preset distance is 100 meters, then the predicted trajectories of the current two flight trajectories are related; otherwise, they are not related.

[0038] In S4, the predicted trajectories of the two previous and next flight trajectories are averaged to obtain a final predicted trajectory, and the two previous and next flight trajectories are connected to the final predicted trajectory.

[0039] like Figure 3 As shown, in this embodiment, there is a number ID 1 、ID 2 、ID 3 Three flight trajectories of ID 1 、ID 2 、ID 3 The endpoints of the three flight trajectories are at the boundary of the monitoring blind area. 1 Do the backward trajectory prediction to get the first backward prediction trajectory For the second flight track ID 2 Make a forward trajectory prediction to get the second forward prediction trajectory For the second flight track ID 2 Do the backward trajectory prediction to get the second backward prediction trajectory For the third flight track ID 3 Make a forward trajectory prediction to get the third forward prediction trajectory The formulas corresponding to the four predicted trajectories are as follows:

[0040]

[0041]

[0042]

[0043]

[0044] Among them, au, bu, cu, and du are the first backward prediction trajectories respectively. Second forward prediction trajectory Second backward prediction trajectory The third forward prediction trajectory The u-th predicted position coordinate point;

[0045] Calculate the first backward prediction trajectory using the Hausdorff distance method With the second forward prediction trajectory Hausdorff distance;

[0046] For the first backward prediction trajectory Each position coordinate point ap in the second forward prediction trajectory The distances ‖ap-bq‖ between the coordinate points bq closest to this point ap in the set are sorted, and then the maximum value of the distance is taken as Gather to One-way Hausdorff distance of a set The specific formula is as follows:

[0047]

[0048] Then for the second forward prediction trajectory Each point in bq is forward-predicted to the first backward-predicted trajectory The distances ‖bq-ap‖ between the points ap closest to this point bq are sorted, and then the maximum value of the distance is taken as Gather to One-way Hausdorff distance of a set The specific formula is as follows:

[0049]

[0050] Pick The maximum of the two The Hausdorff distance H

[0051] The smaller the value, the The more relevant. The smaller the value, the The more relevant.

[0052] If ID 1 Backward trajectory prediction ID 2 The forward trajectory prediction Related, then ID 1 and ID 2 The two trajectories come from the same drone; if ID 2 Backward trajectory prediction ID 3 The forward trajectory prediction Related, then ID 2 and ID 3 The two trajectories come from the same drone. Final ID 1 、ID 2 、ID 3 All three trajectory curves come from the same drone.

[0053] For two correlated prediction trajectories The new trajectory is obtained by the following formula

[0054] For two correlated prediction trajectories The new trajectory is obtained by the following formula

[0055]

[0056] Multiple drone trajectories are connected to obtain a complete trajectory ID as shown in the following formula:

[0057]

[0058] S5: Repeat S4 to judge the two adjacent flight trajectories remaining in the short term, obtain the continuous flight trajectory of the drone in the current drone monitoring space in the short term, and thus obtain the flight trajectory of the drone crossing the monitoring blind spot; then use the corresponding drone control area as the starting point-end point area of ​​the drone to obtain the complete flight trajectory of the drone in the current drone monitoring space.

Claims

1. A method for determining the correlation of UAV trajectories in monitoring blind areas, characterized in that: The following steps are involved: S1: According to the preset time interval, the historical flight data of the UAV obtained by monitoring in the UAV monitoring space is segmented to obtain the corresponding historical flight data segments, then the endpoints of each historical flight data segment are taken to form a historical flight data endpoint set, and then the historical flight data endpoint set is clustered to fit the monitoring blind area boundary of the current UAV monitoring space; S2: In the current drone monitoring space, monitor and obtain the drone flight data in the short term, segment the drone flight data in the short term at preset time intervals to obtain the corresponding short-term flight data segments, and then perform trajectory fitting on each short-term flight data segment to obtain each flight trajectory of the drone in the short term; S3: Take the endpoints of each segment of the UAV's flight trajectory in a short period of time and form a short-term flight data endpoint set, cluster the short-term flight data endpoint set, cluster to obtain the UAV control area, and at the same time, judge whether the endpoints of each segment of the UAV's flight trajectory in a short period of time are within the monitoring blind zone boundary according to the monitoring blind zone boundary of the current UAV monitoring space; S4: For two adjacent flight trajectories in a short period of time, if the endpoints of the two adjacent flight trajectories in a short period of time are both at the monitoring blind zone boundary, the trajectory of the current two adjacent flight trajectories is predicted to obtain the corresponding predicted trajectory, and then the correlation of the predicted trajectories of the current two adjacent flight trajectories is determined; if the predicted trajectories are related, it is considered that the current two adjacent flight trajectories are from the same UAV; then the trajectories of the current two adjacent flight trajectories are connected according to the predicted trajectories of the current two adjacent flight trajectories; if the predicted trajectories are not related, it is considered that they are not from the same UAV; if the endpoints of the two adjacent flight trajectories in a short period of time are not at the monitoring blind zone boundary, no processing is performed; S5: Repeat S4 to judge the two adjacent flight trajectories remaining in the short term, and obtain the continuous flight trajectory of the drone in the current drone monitoring space in the short term; then use the corresponding drone control area as the starting point-end point area of ​​the drone to obtain the complete flight trajectory of the drone in the current drone monitoring space.

2. The method for determining the correlation of UAV trajectories in a monitoring blind area according to claim 1 is characterized in that: In the S1, the endpoint set of historical flight data is clustered to obtain multiple historical clusters, the x-axis direction is set to the horizontal direction, the center point coordinates of the multiple historical clusters are projected to the x-axis, the two projection points x0 and x1 with the farthest distance on the x-axis are selected and recorded as the first and second projection points, the line segment connecting the center point coordinates of the historical clusters corresponding to the first and second projection points is used as the diameter and the first sphere S0 is made with the center of the current line segment as the sphere center, and the outer spherical surface of the first sphere S0 is used as the maximum boundary of the monitoring blind area; Next, a straight line l is made that passes through the center of the first sphere S0 and is perpendicular to the x-axis. Two projection points x2 and x3 that are closest to the straight line l are found on the x-axis on both sides of the straight line l and are recorded as the third and fourth projection points. The line segment connecting the coordinates of the center points of the historical clusters corresponding to the third and fourth projection points is used as the diameter and a second sphere S1 is made with the center of the current line segment as the center. The outer surface of the second sphere S1 is used as the minimum boundary of the monitoring blind spot; the annular area between the maximum boundary and the minimum boundary of the monitoring blind spot is used as the boundary of the monitoring blind spot.

3. The method for determining the correlation of UAV trajectories in a monitoring blind area according to claim 1 is characterized in that: The S3 is specifically: S31: Filter each segment of the UAV's flight trajectory in the short term through the Kalman inverse filter to obtain a discrete flight data sequence corresponding to each segment of the flight trajectory, take the endpoint data of each discrete flight data sequence and form a short-term flight data endpoint set; S32: Clustering the endpoint set of short-term flight data to obtain multiple short-term clusters, taking the space where the short-term clusters formed by clustering the endpoint data at the take-off point and landing point of the drone are located as the drone control area, and judging whether the center points of the remaining short-term clusters other than the drone control area are in the monitoring blind area boundary; If the center point of each short-term cluster is in the monitoring blind zone boundary, then the endpoint data of the current short-term cluster are all in the monitoring blind zone boundary, and the endpoint data of the current short-term cluster are all in the monitoring blind zone boundary, which means that the endpoints of the corresponding flight trajectories are in the monitoring blind zone boundary; otherwise, they are not in the monitoring blind zone boundary.

4. The method for determining the correlation of UAV trajectories in a monitoring blind area according to claim 1, characterized in that: In S4, the Hausdorff distance method is used to calculate the Hausdorff distance between the predicted trajectories of the current two preceding and following flight trajectories. If the Hausdorff distance between the predicted trajectories of the current two preceding and following flight trajectories is less than the preset distance, the predicted trajectories of the current two preceding and following flight trajectories are correlated; otherwise, they are not correlated.

5. The method for determining the correlation of unmanned aerial vehicle trajectories in a monitoring blind area according to claim 1, characterized in that: The preset time intervals in S1 and S2 are the same.

6. The method for determining the correlation of UAV trajectories in a monitoring blind area according to claim 1, characterized in that: In S2, a Kalman filter is used to perform trajectory fitting on each short-term flight data segment.

7. The method for determining the correlation of UAV trajectories in a monitoring blind area according to claim 1, characterized in that: In S4, the predicted trajectories of the two current flight trajectories are averaged to obtain a final predicted trajectory, and the two current flight trajectories are connected to the final predicted trajectory.

Images