A method for real-time high-precision trajectory fusion and connection of a radar video of an intersection traffic scene

Abstract

The application discloses a kind of intersection traffic scene radar video real-time high-precision trajectory fusion method of linking, comprising the following steps: S1: according to the set calculation frequency, obtain the target trajectory point set detected by radar and visual perception equipment;S2: according to the set calculation frequency, obtain the candidate trajectory point set detected by radar and visual perception equipment;S3: the target trajectory point set and candidate trajectory point set detected by radar and visual perception equipment are filtered, and the target trajectory point set and candidate trajectory point set after deduplication are obtained;S4: the position where radar target trajectory is located after deduplication is judged, and radar target trajectory and video candidate trajectory are connected according to the judgment result;S5: the position where video target trajectory is located after deduplication is judged, and video target trajectory and radar candidate trajectory are connected according to the judgment result;S6: the connection of all radar target trajectory and video target trajectory is completed, and complete trajectory set is obtained;Trajectory is divided into two types of target and candidate, and complete vehicle trajectory is obtained by the bidirectional linking of radar and video in the entrance lane stop line and the matching range of exit lane, with high accuracy, fast matching, and effectively preventing jump.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent transportation, in particular to a method for real-time high-precision trajectory fusion and connection of radar and video in intersection traffic scenes. BACKGROUND

[0002] The detection and management of road traffic safety rely on advanced sensor detection technology. Radar has the advantages of all-weather, high real-time, long distance, etc. It is more suitable for detecting traffic scenes with high vehicle speed, relatively low complexity, and long detection range on the road section. Visual perception devices are sensitive to light and have certain decline in detection effect at night. The detection distance is relatively large, generally about 100-200 meters, and is more suitable for installation at intersections for intersection traffic scene detection. Therefore, a holographic simulation solution of road section (radar) plus intersection (vision) is formed. In such a detection scene, how to fuse and connect the vehicle trajectories of radar and vision based on the monitoring information of radar and vision to obtain the complete trajectory of the road vehicle is a problem that needs to be solved in the field of holographic scene application of radar and vision.

[0003] For example, Chinese patent CN114724392A, published on July 8, 2022, discloses a method for dynamic signal control of expressway exit ramp and adjacent intersection. The method includes obtaining traffic flow data, extracting radar and vision fusion trajectory data, constructing and matching trajectories, predicting vehicle trajectories at adjacent intersections, and generating an optimized intersection signal timing strategy for the next cycle. First, the ground lane and high-precision vehicle trajectory at the ramp exit are extracted and fused by a neural network and a re-identification algorithm based on radar and video collection. Then, a generative adversarial network is used to predict the vehicle trajectory at the adjacent intersection in the next phase. Finally, the macro and micro features of the intersection vehicle trajectory are extracted and input into a multi-layer Q reinforcement learning network considering the motion wave and the lane changing congestion of the exit ramp vehicle. The next cycle intersection signal timing strategy is generated by online training to optimize the signal timing of the expressway exit ramp adjacent intersection, improve the intersection traffic volume, and alleviate congestion problems. However, the radar and vision fusion achieved by the method of the present application does not fully consider all cases of radar trajectory data and video trajectory data before fusion, which may cause errors in the final fused trajectory data. SUMMARY

[0004] The technical problem to be solved by the present application is that the existing technology cannot perfectly realize the perfect fusion of radar trajectory data and video trajectory data. A method for real-time high-precision trajectory fusion and connection of radar and video in intersection traffic scenes is proposed, which considers all possible cases of radar trajectory data and video trajectory data, and finally realizes the fusion of all trajectory data to obtain the most perfect fusion data.

[0005] To solve the above technical problems, the technical scheme adopted by the present application is: a method for real-time high-precision trajectory fusion and connection of radar video in intersection traffic scenarios, comprising the following steps:

[0006] S1: According to the set frequency, obtain the target trajectory point set detected by the radar and visual perception device;

[0007] S2: According to the set frequency, obtain the candidate trajectory point set detected by the radar and visual perception device;

[0008] S3: Screen the target trajectory point set and the candidate trajectory point set detected by the radar and visual perception device to obtain the de-duplicated target trajectory point set and the de-duplicated candidate trajectory point set;

[0009] S4: Determine the position of the de-duplicated radar target trajectory, and connect the radar target trajectory and the video candidate trajectory according to the determination result;

[0010] S5: Determine the position of the de-duplicated video target trajectory, and connect the video target trajectory and the radar candidate trajectory according to the determination result;

[0011] S6: Complete the connection of all radar target trajectories and video target trajectories to obtain a complete trajectory set.

[0012] A method for real-time high-precision trajectory fusion and connection of radar video in intersection traffic scenarios first defines the road section range, intersection range and exit lane matching range, and installs the installation points of the radar and visual perception device, then sets various parameters, then obtains the latest target trajectory point set and candidate trajectory point set of the radar and video, then determines the position of each trajectory, and connects the trajectories that meet the conditions, and finally obtains the complete trajectory set of each target vehicle.

[0013] As a preferred embodiment, the step S4 comprises the following contents: the actual time is set as t, the timestamp in the calculation is t'=t-ΔT, if the latest point θ i of the radar target trajectory in [t', t'-T] is inside the intersection, then the time t i of the current latest point trajectory point θ i of the radar target trajectory is set as the time when this piece of radar target trajectory enters the intersection, set the entrance of this piece of target trajectory as entrance i , the lane number before the exit of the road section as d i , and set the radar-visual matching number m=0; if the latest point δ j of the video candidate trajectory in [t'-2T, t'-T] does not enter the intersection, then the time t jThe time when the candidate trajectory enters the intersection, and the time when the candidate trajectory exits the entrance, are considered the entry point. j The lane number before exiting the road section was d. j Let the number of times the radar sight matches n = 0; let the number of times the radar sight matches m ∈ [0, M]. max The radar target trajectory entering the intersection is matched with all video candidate trajectories entering the intersection. If the radar target trajectory and the video candidate trajectory match successfully, the radar target trajectory and the video candidate trajectory are connected. The latest point of the video candidate trajectory within [t′, t′-T] is selected for connection and output to the front end. Due to the time required for image recognition and data transmission between the two types of visual perception devices, the visual trajectory data has a certain delay. To maintain the time synchronization consistency between the two types of detection devices, a delay parameter ΔT is set for the radar detection trajectory data. Assuming the current actual time is t, the timestamp of the radar and video detection data actually involved in the algorithm calculation is t′=t-ΔT. When connecting the radar target trajectory and the video candidate trajectory, it is necessary to first determine the location of the radar target trajectory and the location of the video candidate trajectory. The radar target trajectory that has entered the intersection and has an appropriate number of radar-visual matching times is matched with the video candidate trajectory. After successful matching, the two are connected to obtain the complete trajectory data.

[0014] Preferably, the set of video candidate trajectories currently entering the intersection must satisfy the following condition: the latest point δ of the video candidate trajectory within the time period [t′, t′-T] j =[t j LonQ j LatQ j Inside the intersection; the total number of N trajectory points of the video candidate trajectory on the road segment. video N should be satisfied video >N min , where N min These are configuration parameters. Only video candidate trajectories that meet all the above conditions can be matched with radar target trajectories.

[0015] Preferably, a successful match between the radar target trajectory and the video candidate trajectory requires the following conditions to be met: the two trajectories exit the same entrance; the difference in lane numbers between the two trajectory segments is less than or equal to 1, and both lanes are entrance lanes; the time difference between the two trajectories entering the intersection is less than a certain set threshold T. diff For a radar target trajectory to successfully match a video candidate trajectory, three conditions must be met: entrance i =entrance j 、|d i -d j |≤1 and t i -t j ≤T diff Tdiff With M max For configuration parameters.

[0016] Preferably, when multiple video candidate trajectories meet the matching criteria, the main video trajectory with the same lane number is selected first; when multiple main trajectories have the same priority, the t-track is selected first. i ≤t j The main trajectory of the video, where t j t represents the entry time at the intersection of the main video trajectory. i The radar's main trajectory entry time at the intersection is used; if multiple main trajectories of the same priority still exist, the main trajectory video of the intersection with the smallest entry time difference is selected, i.e., |t i -t j Matching is performed using the minimum possible values. When multiple video candidate trajectories meet the matching criteria, priority must be determined, and matching is performed based on the priority level.

[0017] As a preferred option, if the radar target trajectory fails to match the video candidate trajectory, m = m + 1, n = n + 1: if m < M max If n > M, the radar target trajectory will still be retained as the target trajectory, and the latest trajectory point output to the front end will be the latest trajectory point within the time interval [t′, t′-T], ending this round of matching; max If no radar trajectory matches the current video candidate trajectory, the current video candidate trajectory is considered a new video target trajectory. This trajectory is then removed from the video candidate trajectory set. The latest point of this trajectory within [t′, t′-T] is selected, connected, and output to the front end. If n≤M max If the radar target trajectory fails to match the video candidate trajectory, the final processing method will be determined based on the number of radar-visual matching attempts for both the radar target trajectory and the video candidate trajectory.

[0018] Preferably, step S5 includes the following: if the latest point θ of the video target trajectory within the time period [t′, t′-T] i =[t i LonQ i LatQ i If the target trajectory is within the exit lane matching range, it is determined that the video target trajectory has entered the exit lane; if the target trajectory has not entered the exit lane in the time interval [t′-2T, t′-T], the radar matching count m = 0; if the radar candidate trajectory has not entered the intersection in the time interval [t′-2T, t′-T], its radar matching count n = 0; the radar matching count m ∈ [0, M] is set to [0, M]. maxvideo target track in the range of the exit lane, and all radar candidate tracks in the range of the exit lane are matched; in the latest N sample points, the spatiotemporal parameters of each point of the video target track are matched with the spatiotemporal parameters of each point of the radar candidate track the spatiotemporal parameters of each point of the radar candidate track are traversed point matching is performed, if the matching is successful, the matched point of the candidate track is marked as matched, if the matching is unsuccessful, the first point of the target track is matched with the next point of the candidate track, and the matching is performed until the distance range is met; if the video target track and the radar candidate track are matched successfully, track connection is performed, the latest point of the radar candidate track in [t', t'-T] is selected for connection and output to the front end. After the matching and connection of the radar target track and the video candidate track are completed, the matching and connection of the video target track and the radar candidate track are performed, thereby effectively improving the final track matching accuracy.

[0019] Preferably, the point matching needs to meet the following conditions: the timestamp corresponding to the point of the radar candidate track is within , T0 is a configuration parameter, is the timestamp corresponding to the point of the target track being matched; the difference between the longitude and latitude of the point of the radar candidate track and the longitude and latitude of the point of the video target track is within the range of ΔLON and ΔLAT, wherein ΔLON and ΔLAT are configuration parameters; the angle between the vector formed by the latest N moving sample points of the candidate track and the vector formed by the latest N moving sample points of the target track should be less than A1, wherein A1 is a configuration parameter. If the candidate point and the point of the target track being matched meet the above three conditions at the same time, it is considered that the two points are matched successfully, the matching is started from the latest point of the video target track and the latest point of the radar candidate track, if the matching is successful, the matched point of the candidate track is marked as matched, if the matching is unsuccessful, the first point of the target track is matched with the next point of the candidate track, and the matching is performed until the distance range is met, the point matching of the N target tracks is performed, if the number of matched points is < M min (configuration parameter, default is 2) points, the candidate track is not matched successfully, if ≥ M min points, the candidate track is matched successfully. The connection process is entered, if multiple candidate samples are matched, the candidate track closest to the last point of the target track is selected for connection.

[0020] Preferably, if the video target track and the radar candidate track are not matched successfully, m = m + 1, n = n + 1: if m > M maxIf m>M, it is considered that the video target track has no radar candidate track data matched therewith, the matching of the video target track is terminated, and the radar candidate track is no longer output to the front end. max If n>M, it is determined that there is no video target track matched with the radar candidate track, it is determined that the radar candidate track is a new radar target track, the track is deleted from the radar candidate track set, and the latest point of the radar candidate track in [t', t'-T] is selected for splicing and output to the front end. max If n>M, it is determined that there is no video target track matched with the radar candidate track, it is determined that the radar candidate track is a new radar target track, the track is deleted from the radar candidate track set, and the latest point of the radar candidate track in [t', t'-T] is selected for splicing and output to the front end. max If n>M, it is determined that there is no video target track matched with the radar candidate track, it is determined that the radar candidate track is a new radar target track, the track is deleted from the radar candidate track set, and the latest point of the radar candidate track in [t', t'-T] is selected for splicing and output to the front end. If the video target track and the radar candidate track fail to match, the final processing mode of the video target track and the radar candidate track is determined according to the radar-visual matching times of the video target track and the radar-visual matching times of the radar candidate track.

[0021] Preferably, the radar is arranged on a road section, the visual perception device is arranged at an intersection, a stop line is arranged at the boundary between the road section and the intersection, and the exit lane stop line and the 30 meters after the exit lane stop line are the exit lane matching range. When the road section range, the intersection range and the exit lane matching range are defined, the boundary between the road section range and the intersection range is defined according to the stop line therebetween, and the exit lane matching range is the exit lane stop line and the 30 meters after the exit lane stop line.

[0022] The substantial effect of the present application is that the present application designs a radar-video real-time high-precision track fusion splicing method for an intersection traffic scene, the track is divided into target and candidate types by means of the channelization structure of the intersection and the auxiliary calibration range at the exit lane, the bidirectional splicing of the radar and the video is performed in the entrance lane stop line and the exit lane matching range, the complete vehicle track of the road section-intersection-road section is obtained, the accuracy is high, the matching is fast, and the jump is effectively prevented; and the holographic of the urban road is also expanded, which can be from point to line, from line to surface, and form a holographic simulation at the road network level. BRIEF DESCRIPTION OF DRAWINGS

[0023] Figure 1 The step flowchart of the first embodiment;

[0024] Figure 2 The radar and visual perception device layout diagram of the first embodiment;

[0025] Figure 3 The exit lane matching range of the first embodiment.

[0026] 1, visual perception device, 2, radar, 3, exit lane matching range. DETAILED DESCRIPTION

[0027] The specific embodiments of the present application will be further described in detail below with specific examples and in conjunction with the accompanying drawings.

[0028] Example one:

[0029] A method for real-time high-precision trajectory fusion and connection of intersection traffic scene radar video, as shown in Figure 1 , a road section multi-radar detection area trajectory connection optimization method can be seen, and the operation process is as follows: step one, select radar, video layout points, and set the radar device delay time. Define the road section range, the range in the intersection, and the matching range of the exit lane. Complete the setting of each parameter. Step two: filter the target trajectory set to obtain the de-duplicated radar / video target trajectory set; step three: filter the candidate trajectory set to obtain the de-duplicated radar / video candidate trajectory set; step four: judge whether the de-duplicated radar target trajectory is located in the intersection range, and connect with the candidate trajectory detected by the video; step five: judge whether the de-duplicated video target trajectory is located in the exit lane matching range, and connect with the candidate trajectory detected by the radar; step six: traverse all trajectories in the radar / video target trajectory set to obtain the complete trajectory set of each vehicle. Through the bidirectional connection of radar and video in the entrance stop line and the exit matching range, the complete vehicle trajectory of road section-intersection-road section is obtained.

[0030] The specific embodiments of the present application will be further described in detail below with specific examples and in conjunction with the accompanying drawings.

[0031] Taking an intersection in A city as an example, the parameter table used in example one is shown in the following table:

[0032]

[0033] Step 1: Select radar 2 and video layout points. As shown in Figure 2 , millimeter wave radars are deployed on the exit lane traffic light poles of each entrance, facing the entrance lane of the opposite entrance. The electric police and other visual perception devices 1 are deployed on the electric police poles of the entrance lane of each entrance, facing the exit lane of the opposite entrance, while ensuring that the video detection range includes about 20 meters behind the entrance stop line. According to the delay of the on-site video data and the intersection flow, set the radar device delay time ΔT and the calculation frequency T.

[0034] Step 2: Define the road section range, the range in the intersection, and the matching range of the exit lane 3, and the matching range is calibrated as Figure 3 .

[0035] Step 3: Filter the target trajectory set to obtain the de-duplicated radar / video target trajectory set, and the specific steps are as follows:

[0036] Step 31 sets the current time as t, the device delay time ΔT, and the calculation frequency as T. The selected radar / video track data time period for this round of calculation is [t', t'-T], where t'=t-ΔT.

[0037] Step 32, the target track is the radar / video detection track actually output to the front end for holographic display. The latest detected target track point sets of radar and video are obtained and de-duplicated to obtain the final radar / video target track set.

[0038] Step 4 filters the candidate track set to obtain the de-duplicated radar / video candidate track set. The specific steps are as follows:

[0039] Step 41 sets the current time as t, the device delay time ΔT, and the calculation frequency as T. The selected radar / video track data time period for this round of calculation is [t', t'-T], where t'=t-ΔT.

[0040] Step 42, the candidate track is the radar / video detection track, which is not output to the front end for holographic display when it is not connected by the target track. The latest detected candidate track point sets of radar and video are obtained and de-duplicated to obtain the final radar / video candidate track set.

[0041] Step 5 judges whether the de-duplicated radar target track is located within the intersection range and is connected with the candidate track detected by the video. The specific steps are as follows:

[0042] Step 51 judges whether the radar target track enters the intersection

[0043] If the latest point θ i of the target track vehicle_id1 in the time period [t', t'-T] is inside the intersection, it is judged that the radar target track vehicle_id1 enters the intersection. i = [t i , LonQ i , LatQ i ]

[0044] Step 52, if the target track vehicle_id1 did not enter the intersection at the last time period, i.e., [t'-2T, t'-T], the time t i of the latest point track point θ i is set as the time when the radar target track enters the intersection. Set the entrance of the target track as entrance i , the lane number before the exit section as d i , and the radar / video matching times m=0.

[0045] Step 53 filters the video candidate track set currently entering the intersection, which needs to meet the following conditions:

[0046] The latest point δ of the video candidate trajectory video_id1 in the time period [t', t'-T] j = [t j , LonQ j , LatQ j ] is inside the intersection; the total number of trajectory points N of video_id1 on the road segment video Should satisfy N video > N min ;

[0047] Step 54 If the video candidate trajectory video_id1 does not enter the intersection in the last time period, i.e., [t'-2T, t'-T], set the time t j of the latest point δ j as the time when the candidate trajectory enters the intersection, the exit of the candidate trajectory is entrance j , the lane number before the candidate trajectory exits the road segment is d j , and set n=0.

[0048] Step 55 Radar video trajectory matching connection: for all radar target trajectories entering the intersection with radar video matching times m∈[0, M max ], match them with all video candidate trajectories entering the intersection, set the radar target trajectory as vehicle_id1 and the video candidate trajectory as video_id1, and the matching success condition is:

[0049] The two trajectories exit the same entrance, i.e., entrance i = entrance j ;

[0050] The difference between the lane numbers of the two trajectories is less than or equal to 1, and both lanes are entrance lanes, i.e., |d i -d j |≤1;

[0051] The time difference between the two trajectories entering the intersection is less than a certain threshold t i -t j ≤T diff ;

[0052] Step 56 If there are multiple video candidate trajectories that meet the matching conditions, then:

[0053] Preferentially select the video main trajectory with the same lane number, i.e., d i =d j ;

[0054] If there are still multiple main trajectories of the same priority, preferentially select t i ≤tj video main track (where t j is the video main track intersection entry time, t i is the radar main track intersection entry time);

[0055] If there are still multiple main tracks of the same priority, the video main track with the smallest difference in intersection entry time, i.e., |t i -t j | is selected for matching;

[0056] Step 57 If the radar target track vehicle_id1 matches the video candidate track video_id1 successfully, track splicing is performed, the latest point of the video candidate track within [t', t'-T] is selected for splicing and output to the front end, and the matching and splicing are successful. The radar target track video_id1 is deleted from the radar target track set, the video candidate track vehicle_id1 becomes a video target track, and the matching and splicing of the current main track are ended.

[0057] Step 58 If the radar target track vehicle_id1 fails to match the video candidate track video_id1, m = m + 1 and n = n + 1:

[0058] If m > M max , it is considered that there is no video track data matching the radar target track, and the track is terminated. The radar track where the track is located is no longer output to the front end, but is kept as a radar candidate track in the intersection for splicing.

[0059] If m < M max , the radar target track is still kept as a target track, and the track point output to the front end is the latest track point within the time [t', t'-T], and the current matching is ended.

[0060] If n > M max , it is determined that there is no radar track matching the video candidate track, and the video candidate track is determined to be a new video target track. The track is deleted from the video candidate track set, and the latest point of the track within [t', t'-T] is selected for splicing and output to the front end.

[0061] If n ≤ M max , the track is still kept as a video candidate track and is not output to the front end, and waits for splicing with the radar target track, and the current matching is ended.

[0062] Step 6 It is determined whether the de-duplicated video target track is located within the exit lane range and is spliced with the candidate track detected by the radar. The specific steps are as follows:

[0063] Step 61 judges whether the video target track enters the exit lane

[0064] If the latest point θ i = [t i , LonQ i , LatQ i ] of the video target track video-id2 in the time period [t', t'-T] is inside the exit lane matching range, it is judged that the video target track video_id2 enters the exit lane.

[0065] Step 62 sets the radar-video matching times m=0 of the target track video_id2 if the target track video_id2 does not enter the exit lane in the last time period, i.e., [t'-2T, t'-T].

[0066] Step 63 screens the radar candidate track set currently entering the exit lane range, which needs to meet the following conditions: the latest point δ j = [t j , LonQ j , LatQ j ] of the radar candidate track vehicle_id2 in the time period [t', t'-T] is inside the intersection.

[0067] Step 64 sets the radar-video matching times n=0 of the radar candidate track vehicle_id2 if the radar candidate track vehicle_id2 does not enter the intersection in the last time period, i.e., [t'-2T, t'-T].

[0068] Step 65 radar-video track matching connection: for all video target tracks entering the exit lane range with radar-video matching times m∈[0, M max ], the video target track is matched with all radar candidate tracks entering the exit lane range, with the video target track being video_id2 and the radar candidate track being vehicle_id2.

[0069] In the latest N sample points, according to the space-time parameters of each point of the video target track the space-time parameters of each point of the radar candidate track are traversed to perform point matching, which meets the following conditions to be successful point matching:

[0070] 1) the timestamp of the candidate track point is , the timestamp of the target track point corresponding to the current matching;

[0071] 2) the difference between the longitude and latitude of the candidate track point and the longitude and latitude of the target track point is within the range of ΔLON and ΔLAT.

[0072] 3) Direction angle constraint, the angle between the vector formed by the latest N moving sample points of the candidate trajectory (the vector is formed by the first point and the last point of the candidate trajectory) and the vector formed by the latest N moving sample points of the target trajectory should be less than A1;

[0073] If the candidate point is within the distance range of the target trajectory point being matched, and the direction angle constraint is met, the two points are considered to be matched successfully. Starting from the latest point of the video target trajectory and the latest point of the radar candidate trajectory, if the matching is successful, the matched point of the candidate trajectory is marked as matched; if the matching is not successful, the first point of the target trajectory is matched with the next point of the candidate trajectory, and the matching is continued until the distance range is met.

[0074] After matching the N target trajectory points, if the number of matched points is less than M min , the candidate trajectory is not matched successfully; if the number of matched points is greater than or equal to M min , the candidate trajectory is matched successfully. Enter the linking process, and if multiple candidate samples are matched, select the candidate trajectory closest to the last point of the target trajectory for linking.

[0075] Step 67 If the video target trajectory video_id2 and the radar candidate trajectory vehicle_id2 are matched successfully, perform trajectory linking, select the latest point of the radar candidate trajectory within [t', t'-T] for linking and output to the front end, and the matching and linking are successful. Remove the video target trajectory video_id2 from the video target trajectory set, and the radar candidate trajectory vehicle_id2 becomes a radar target trajectory, ending the matching and linking of the current round of the display main trajectory.

[0076] Step 68 If the video target trajectory video_id2 and the radar candidate trajectory vehicle_id2 are not matched, m = m + 1, n = n + 1:

[0077] If m > M max , it is considered that there is no radar trajectory data matched with the video target trajectory, and the trajectory is terminated, and the radar trajectory is no longer output to the front end.

[0078] If m < M max , the video target trajectory is still retained as a target trajectory, and the latest trajectory point within the time [t', t'-T] is output to the front end as the trajectory point, and the matching is ended.

[0079] If n > M maxIf not, it is determined that there is no video track matched with the radar candidate track, and it is determined that the radar candidate track is a new radar target track. The track is deleted from the radar candidate track set, and the latest point of the track within [t', t'-T] is selected to be connected and output to the front end.

[0080] If n≤M max The track is still a radar candidate track, is not output to the front end, and waits to be connected with a video target track, and the matching in this round is ended.

[0081] Step 7: All tracks in the radar / video target track set are traversed to obtain a complete track set of each vehicle.

[0082] The above embodiments only express several embodiments of the present application, and the description is relatively specific and detailed, but it cannot be understood as a limitation on the scope of the patent. It should be noted that, for ordinary skilled persons in the art, several modifications and improvements can be made without departing from the concept of the present application, and these are within the protection scope of the present application.

Claims

1. A method for real-time high-precision trajectory fusion and connection of intersection traffic scene radar video, characterized in that, The method comprises the following steps: S1: acquiring a target track point set detected by a radar and a visual perception device according to a set frequency; S2: acquiring a candidate track point set detected by the radar and the visual perception device according to a set frequency; S3: screening the target track point set and the candidate track point set detected by the radar and the visual perception device to obtain a de-duplicated target track point set and a de-duplicated candidate track point set; S4: judging a position of the de-duplicated radar target track, and connecting the radar target track and the video candidate track according to a judgment result; S5: judging a position of the de-duplicated video target track, and connecting the video target track and the radar candidate track according to a judgment result; S6: completing connection of all radar target tracks and video target tracks to obtain a complete track set.

2. The method of claim 1, wherein, Step S4 includes the following: the actual time is set as t, the timestamp in the calculation is t′=t-ΔT, and if the latest point θ of the radar target trajectory is within [t′,t′-T]... i If the radar target is inside the intersection, its trajectory enters the intersection; if the latest point of the radar target trajectory within [t′-2T, t′-T] is not inside the intersection, then its current latest trajectory point θ is set. i Time t i Let the time when the radar target trajectory enters the intersection be denoted as "entrance". i The lane number before exiting the road section was d. i Let the number of matching attempts m = 0; if the video candidate trajectory does not enter the intersection at time [t′-2T, t′-T], then set its latest point δ. j Time t j The time when the candidate trajectory enters the intersection, and the time when the candidate trajectory exits the entrance, are considered the entry point. j The lane number before exiting the road section was d. j Let the number of times the radar sights match is n = 0; The radar target trajectory of the entering intersection with the number of radar matching times m e [0, M max ] is matched with all video candidate trajectories of the entering intersection; If the radar target track and the video candidate track are matched successfully, the radar target track and the video candidate track are connected, the latest point of the video candidate track in [t', t'-T] is selected for connection, and the connection result is output to a front end.

3. The method of claim 2, wherein, The video candidate trajectory set of the current entering intersection, needs to meet the following conditions: the latest point δ of the video candidate trajectory in the time period [t', t'-T] j = [t j , LonQ j , LatQ j ] is inside the intersection; the total number of trajectory points N of the video candidate trajectory on the road segment video should meet N video > N min , wherein N min is a configuration parameter.

4. The method of claim 2 or 3, wherein, The radar target trajectory and the video candidate trajectory match successfully need to meet the following conditions: the two trajectories exit the entrance; the difference between the two trajectory lane numbers is less than or equal to 1, and the lanes are both entrance lanes; the time difference between the two trajectories entering the intersection is less than a certain set threshold T diff .

5. The method of claim 2 or 3, wherein, When multiple video candidate trajectories meet the matching criteria, the main video trajectory with the same lane number is selected first; if multiple main trajectories have the same priority, the t-track is selected first. i ≤t j The main trajectory of the video, where t j t represents the entry time at the intersection of the main video trajectory. i This refers to the radar's main trajectory entry time at the intersection; If there are still multiple main tracks of the same priority, the main track of the intersection video with the minimum time difference of driving into the intersection, i.e. |t i -t j |minimumis matched.

6. The method of claim 2 or 3, wherein, If the radar target trajectory and the video candidate trajectory fail to match, m = m + 1, n = n + 1: if m < M max , the radar target trajectory is still retained as a target trajectory, and the latest trajectory point in the [t', t'-T] time is output to the front-end trajectory point, and the matching in this round is ended; if n > M max , it is judged that there is no radar trajectory that can match the video candidate trajectory, the video candidate trajectory is a new video target trajectory, the trajectory is deleted from the video candidate trajectory set, and the latest point of the trajectory in [t', t'-T] is selected to be connected and output to the front-end; if n ≤ M max , the trajectory is still a video candidate trajectory, and is not output to the front-end, and waits to be connected with the radar target trajectory, and the matching in this round is ended.

7. The method of claim 1, wherein, The step S5 comprises the following: if the latest point θ i = [t i , LonQ i , LatQ i ] of the video target track in the time period [t', t'-T] is inside the exit lane matching range, it is determined that the video target track enters the exit lane; If the target trajectory does not enter the exit lane at [t'-2T, t'-T], set the radar-visual matching times m=0; if the radar candidate trajectory does not enter the intersection at [t'-2T, t'-T], set its radar-visual matching times n=0; match the video target trajectory with the entering exit lane range of the radar-visual matching times m∈[0, M max ] with all the radar candidate trajectories entering the exit lane range; in the latest N sample points, according to the space-time parameters of each point of the video target trajectory traverse the space-time parameters of each point of the radar candidate trajectory perform point matching, if the matching is successful, the matched point of the candidate trajectory is marked as matched, if the matching is not successful, the first point of the target trajectory is matched with the next point of the candidate trajectory, and the matching is performed until the distance range is met; if the video target trajectory and the radar candidate trajectory are matched successfully, perform trajectory splicing, select the latest point of the radar candidate trajectory within [t', t'-T] for splicing and output to the front end.

8. The method of claim 7, wherein, The point position matching needs to meet the following conditions: the time stamp corresponding to the radar candidate track point position is in T0 is a configuration parameter, is the time stamp corresponding to the target track point position currently being matched. the longitude and latitude of the radar candidate track point the longitude and latitude of the video target track point the difference between the longitude and latitude of the radar candidate track point and the longitude and latitude of the video target track point is within a range of ΔLON and ΔLAT, where ΔLON and ΔLAT are configuration parameters; the angle between the vector formed by the latest N moving sample points of the candidate track and the vector formed by the latest N moving sample points of the target track should be less than A1, where A1 is a configuration parameter.

9. The method of claim 7 or 8, wherein, If the video target track and the radar candidate track matching fails, m = m + 1, n = n + 1: if m > M max It is considered that the video target track has no matching radar candidate track data, and the matching of the video target track is terminated, and the radar candidate track is no longer output to the front end. If m < M max , the video target track is kept as a video target track, the latest track point in [t', t'-T] is output to the front end, and the current matching is ended; if n > M max , it is determined that there is no video target track that can be matched with the radar candidate track, it is determined that the radar candidate track is a new radar target track, the track is deleted from the radar candidate track set, and the latest point of the radar candidate track in [t', t'-T] is selected to be connected and output to the front end; if n ≤ M max , the track still serves as a radar candidate track, is not output to the front end, and waits to be connected with a video target track, and the current matching is ended.

10. The method of claim 1, wherein, The radar is arranged on a road section, the visual perception device is arranged on an intersection, a stop line is arranged at a boundary between the road section and the intersection, and a stop line at an exit lane of the intersection and 30 meters after the stop line are an exit lane matching range.

Images