A method for automatic site selection of base stations based on environment perception

Abstract

The present invention relates to an automatic site selection method for a perception base station based on an environment, comprising the following steps: S1, a perception base station receives an automatic site selection instruction and starts a site selection program; S2, obtains pixel coordinate data of a target perception area and obtains a target image corresponding to the pixel coordinate data; S3, the perception base station rotates a camera and shoots an image through a detector posture adjustment device, and when the target image is located in the middle of the shot image, obtains a current image shot by the camera and a current point cloud collected by a laser radar; S4, performs image recognition on the current image to obtain environmental features; S5, based on the environmental features and point cloud features obtained in step S, determines whether the current position of the perception base station satisfies preset conditions at the same time; S6, generates and executes a movement instruction, triggers the perception base station to move, and executes steps S1 to S5 after the perception base station moves until the site selection program ends.

Description

Technical Field

[0001] The present invention belongs to the technical field of roadside laser radar site selection, and relates to an automatic site selection method for a mobile roadside perception base station based on an environment. Background Art

[0002] The vehicle-mounted LiDAR perception system is a system that integrates three technologies: LiDAR, Global Positioning System GPS, and Inertial Measurement Unit IMU. It has the advantages of high resolution, good confidentiality, and strong anti-interference ability. In smart cars, LiDAR is used for map drawing, positioning, obstacle detection, etc. With the rapid development of LTE-V2X vehicle networking, how to deploy LiDAR on the roadside has attracted more and more attention.

[0003] The deployment plan of existing roadside LiDARs is usually completed on-site by staff based on experience. After the deployed LiDAR is tested, the staff will go to the site to adjust or optimize it. For example, the "A roadside LiDAR deployment position assessment system and method" disclosed in Chinese invention patent application CN115685221A obtains an accurate LiDAR installation position assessment index based on point cloud quality assessment. A large amount of data collection and cleaning is required during testing, and then model training and result evaluation are performed based on the cleaned data, which is extremely time-consuming and labor-intensive. Therefore, how to complete the optimization of LiDAR deployment in a short period of time and at a low cost has become an urgent problem to be solved. Summary of the invention

[0004] In response to the above technical problems, the purpose of the present invention is to provide an automatic site selection method for a perception base station based on the environment, which collects road images and point cloud data through cameras and lidars, extracts scene features and point cloud features of the current scene based on the road images and point cloud data, and determines whether the current position of the base station meets preset conditions based on the scene features and point cloud features. If not, a moving instruction is generated and executed to trigger the movement of the perception base station.

[0005] In order to achieve the above object, the present invention provides the following technical solutions:

[0006] A method for automatic site selection of a perception base station based on an environment, wherein the perception base station comprises a mobile body, a switch, an edge processor, a camera, a laser radar, a detector posture adjustment device and a site selection client; wherein the switch and the edge processor are placed in a cavity of the mobile body, and the camera and the laser radar are installed in parallel on the mobile body through the detector posture adjustment device; the edge processor is electrically connected to the switch, the detector posture adjustment device, the camera and the laser radar respectively; the detector posture adjustment device can accurately adjust the horizontal height, horizontal angle and pitch angle of the camera and the laser radar; and the site selection client is communicatively connected to the perception base station.

[0007] The automatic site selection method comprises the following steps:

[0008] S1. The perception base station receives an automatic site selection instruction and starts a site selection procedure; the automatic site selection instruction is sent by the site selection client to the perception base station, or is generated by the perception base station itself in response to a user operation; the perception base station controls the camera through the detector posture adjustment device to collect 360° perception data of the current scene, obtains a panoramic image of the current scene, and sends the panoramic image to the site selection client;

[0009] S2. The site selection client obtains pixel coordinate data of the target perception area according to the target perception area delineated by the staff on the basis of the panoramic image, and sends the pixel coordinates of the target perception area to the perception base station; the perception base station receives the pixel coordinate data sent by the site selection client, and obtains the target image corresponding to the pixel coordinate data;

[0010] S3, the sensing base station rotates the camera and captures an image through the detector attitude adjustment device. When the target image is located in the middle of the captured image, the current image captured by the camera and the current point cloud collected by the laser radar are obtained;

[0011] S4, performing image recognition on the current image to obtain environmental features; the environmental features include obstacles, lane types, and traffic markings;

[0012] Rasterize the current point cloud to obtain the number of point clouds p(vi) and the signal-to-noise ratio of each grid in the current image; obtain point cloud features according to the number of point clouds and the signal-to-noise ratio, wherein the point cloud features include the number of grids n whose point cloud number is less than the point cloud threshold, the number of grids m whose signal-to-noise ratio is less than the signal-to-noise ratio threshold, and the overall quality parameter H of the point cloud in the area corresponding to the current image;

[0013] S5. Based on the environmental features and point cloud features obtained in step S4, determine whether the current location of the sensing base station satisfies the following preset conditions at the same time. If the environmental features and point cloud features meet all the preset conditions at the same time, the site selection procedure ends; otherwise, execute step S6 to adjust the location of the sensing base station;

[0014] Preset condition 1: There is no obstacle or the occlusion rate of the obstacle is less than the occlusion rate threshold;

[0015] Preset condition 2: If the lane type is a curve, the lane lines within the preset distance at both ends of the curve fall into the detection area of ​​the camera and lidar;

[0016] Preset condition 3: The clarity of traffic markings is greater than the preset value;

[0017] Preset condition 4: The overall quality parameter H of the point cloud is greater than the quality parameter threshold;

[0018] Preset condition 5: n×λ1+ m×λ2<0.5; n is the number of grids whose point cloud number is less than the point cloud threshold, m is the number of grids whose signal-to-noise ratio is less than the signal-to-noise ratio threshold, λ1, λ2 are preset weights, and λ1+ λ2=1 / N, N is the total number of grids;

[0019] S6. Generate and execute a moving instruction to trigger the movement of the sensing base station. After the sensing base station moves, execute steps S1 to S5 until the site selection procedure is completed.

[0020] In step S4, the point cloud threshold is ,in, P is the average number of point clouds in a grid.

[0021] In step S4, the signal-to-noise ratio threshold is set to 70-90.

[0022] In step S4, the overall quality parameter of the point cloud H The calculation method is as follows:

[0023] According to the importance of the target area corresponding to the current image, the type of road structure, the frequency of traffic target occurrence, and the possibility of pedestrian intrusion, each grid is assigned a corresponding weight. w ( v i ), 0< w ( v i )<1, the sum of all grid weights in the target area is 1, and the number of point clouds for each grid p ( v i ) and its weight, calculate its conditional entropy:

[0024]

[0025] The overall quality parameters of the point cloud in the target area are calculated using the following formula:

[0026]

[0027] in represents the point cloud distribution in the target area, N Represents the sum of all grid numbers in the target area.

[0028] In step S5, the occlusion rate threshold in the preset condition 1 is 10%-25%.

[0029] In the step S5, the preset distance in the preset condition 2 is 20-50m.

[0030] In step S5, the preset value in the preset condition 3 is 0.1.

[0031] In step S5, the quality parameter threshold in the preset condition 4 is set according to the accuracy of point cloud target recognition required by the site selection.

[0032] In the step S6, the mobile roadside sensing base station moves according to the judgment result of the preset conditions in step S5. When the preset condition 1 is not met, the mobile roadside sensing base station moves in the direction close to the obstacle; when the preset condition 2 is not met, the mobile roadside sensing base station moves in the direction away from the target area; when the preset condition 3, the preset condition 4 or the preset condition 5 is not met, the horizontal angle and / or the pitch angle of the base station are fine-tuned.

[0033] Compared with the prior art, the present invention has the following beneficial effects:

[0034] The present invention realizes rapid evaluation of base station site selection by comprehensively considering the environment and point cloud quality, with low cost and high efficiency. The present invention comprehensively considers point cloud quality, point cloud grid signal-to-noise ratio, obstacles, occlusion rate, lane type, lane marking clarity, the relationship between lane and detection area, etc., and calculates relevant parameters based on the target detection area, and does not care about the point cloud quality outside the area. On the one hand, it can effectively reduce the amount of site selection calculation, and on the other hand, it can avoid noise interference in non-target areas; it is a more comprehensive method with more practical installation site selection significance, and the site selection success rate based on this method is close to 100%. DETAILED DESCRIPTION

[0035] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0036] The present invention provides an automatic site selection method for a mobile roadside perception base station based on an environment. An automatic site selection method for a perception base station based on an environment, the mobile roadside perception base station comprises a mobile vehicle body, a switch, an edge processor, a camera, a laser radar, a detector posture adjustment device and a site selection client; wherein the switch and the edge processor are placed in a cavity of the mobile vehicle body, and the camera and the laser radar are installed in parallel on the mobile vehicle body through the detector posture adjustment device; the edge processor is electrically connected to the switch, the detector posture adjustment device, the camera and the laser radar respectively; the detector posture adjustment device can accurately adjust the horizontal height, horizontal angle and pitch angle of the camera and the laser radar; the site selection client is communicatively connected to the mobile roadside perception base station, and is used to receive and display the panoramic image, point cloud data and related parameters of the current scene fed back by the mobile roadside perception base station, so that the staff can delineate the target perception area based on the panoramic image, and send the target perception area delineated by the staff to the mobile roadside perception base station.

[0037] The automatic site selection method comprises the following steps:

[0038] S1. The mobile roadside perception base station receives an automatic site selection instruction and starts a site selection procedure; the automatic site selection instruction is sent by the site selection client to the mobile roadside perception base station, or is generated by the mobile roadside perception base station itself in response to a user operation; the mobile roadside perception base station controls the camera through the detector posture adjustment device to collect 360° perception data of the current scene, obtains a panoramic image of the current scene, and sends the panoramic image to the site selection client;

[0039] S2. The site selection client obtains pixel coordinate data of the target perception area according to the target perception area delineated by the staff on the basis of the panoramic image, and sends the pixel coordinates of the target perception area to the mobile roadside perception base station; the mobile roadside perception base station receives the pixel coordinate data sent by the site selection client, and obtains the target image corresponding to the pixel coordinate data;

[0040] S3, the sensing base station rotates the camera and captures an image through the detector attitude adjustment device. When the target image is located in the middle of the captured image, the current image captured by the camera and the current point cloud collected by the laser radar are obtained;

[0041] S4. Perform image recognition on the current image to obtain environmental features, which include obstacles, lane types, traffic markings, etc. Obstacles refer to objects that block detection objects in the target image except vehicles, pedestrians, traffic guides, cones, etc. Lane type refers to whether the lane belongs to a curve, a crossroads, or a T-junction. Traffic markings include stop lines, prohibition markings, crosswalks, etc.

[0042] The current point cloud is rasterized to obtain the point cloud number p(vi) and the signal-to-noise ratio of each grid in the current image corresponding to the grid; the point cloud features are obtained according to the point cloud number and the signal-to-noise ratio, and the point cloud features include the number of grids n whose point cloud number is less than the point cloud threshold, the number of grids m whose signal-to-noise ratio is less than the signal-to-noise ratio threshold, and the overall quality parameter H of the point cloud in the corresponding area of ​​the current image.

[0043] The point cloud threshold is , where P is the average number of point clouds in a grid.

[0044] The embodiment of the present invention adopts PSNR signal-to-noise ratio, and the signal-to-noise ratio threshold is set to 70-90.

[0045] The calculation method of the point cloud overall quality parameter H is as follows:

[0046] According to the importance of the target area corresponding to the current image, the type of road structure, the frequency of traffic targets, the possibility of pedestrian intrusion, etc., each grid is assigned a corresponding weight w(vi), 0<w(vi)<1, and the sum of all grid weights in the target area is 1. For each grid point cloud number p(vi) and its weight, its conditional entropy is calculated:

[0047]

[0048] The overall quality parameters of the point cloud in the target area are calculated using the following formula:

[0049]

[0050] in represents the point cloud distribution in the target area, and N represents the sum of all grids in the target area. The higher the H value calculated in the end, the higher the point cloud quality in the current target area.

[0051] S5. Based on the environmental features and point cloud features obtained in step S4, determine whether the current location of the sensing base station satisfies the following preset conditions at the same time. If the environmental features and point cloud features meet all the preset conditions at the same time, the site selection procedure ends; otherwise, execute step S6 to adjust the location of the sensing base station;

[0052] Preset condition 1: There is no obstacle or the occlusion rate of the obstacle is less than the occlusion rate threshold; the occlusion rate threshold is 10%-25%;

[0053] Preset condition 2: If the lane type is a curve, the lane lines within a preset distance at both ends of the curve fall into the detection area of ​​the camera and the lidar; the preset distance is 20-50m;

[0054] Preset condition 3: the clarity of the traffic marking is greater than a preset value; different clarity estimation methods have different preset values. In the embodiment of the present invention, blur detection is used, and the preset value is 0.1;

[0055] Preset condition 4: the overall quality parameter H of the point cloud is greater than the quality parameter threshold; the quality parameter threshold is set according to the statistical analysis of the detection data. For example, when the overall quality parameters of the point cloud are H1, H2, and H3, the accuracy of point cloud target recognition is 80%, 90%, and 95% through big data statistics. If the site selection requires the accuracy of point cloud target recognition to be 90%, then the threshold can be set to H2, and so on;

[0056] Preset condition 5: n×λ1+ m×λ2<0.5; n is the number of grids whose point cloud number is less than the point cloud threshold, m is the number of grids whose signal-to-noise ratio is less than the signal-to-noise ratio threshold, λ1, λ2 are preset weights, and λ1+ λ2=1 / N, N is the total number of grids;

[0057] S6. Generate and execute a moving instruction to trigger the movement of the sensing base station. After the sensing base station moves, execute steps S1 to S5 until the site selection procedure is completed.

[0058] Further, in step S6, the sensing base station moves according to the judgment result of the preset condition in step S5. When preset condition 1 is not met, the sensing base station moves in a direction close to the obstacle; when preset condition 2 is not met, the sensing base station moves in a direction away from the target area; when preset condition 3, preset condition 4 or preset condition 5 is not met, the horizontal angle and / or pitch angle of the base station is fine-tuned.

Claims

1. A method for automatic site selection of a perception base station based on an environment, wherein the perception base station comprises a mobile vehicle, a switch, an edge processor, a camera, a laser radar, a detector posture adjustment device and a site selection client; wherein: The switch and edge processor are placed in the cavity of the mobile vehicle body, and the camera and laser radar are installed in parallel on the mobile vehicle body through the detector posture adjustment device; the edge processor is electrically connected to the switch, the detector posture adjustment device, the camera and the laser radar respectively; the detector posture adjustment device can accurately adjust the horizontal height, horizontal angle and pitch angle of the camera and the laser radar; the site selection client is connected to the perception base station in communication, characterized in that The automatic site selection method comprises the following steps: S1. The perception base station receives an automatic site selection instruction and starts a site selection procedure; the automatic site selection instruction is sent by the site selection client to the perception base station, or is generated by the perception base station itself in response to a user operation; the perception base station controls the camera through the detector posture adjustment device to collect 360° perception data of the current scene, obtains a panoramic image of the current scene, and sends the panoramic image to the site selection client; S2. The site selection client obtains pixel coordinate data of the target perception area according to the target perception area delineated by the staff on the basis of the panoramic image, and sends the pixel coordinates of the target perception area to the perception base station; the perception base station receives the pixel coordinate data sent by the site selection client, and obtains the target image corresponding to the pixel coordinate data; S3, the sensing base station rotates the camera and captures an image through the detector attitude adjustment device. When the target image is located in the middle of the captured image, the current image captured by the camera and the current point cloud collected by the laser radar are obtained; S4, performing image recognition on the current image to obtain environmental features; the environmental features include obstacles, lane types, and traffic markings; Rasterize the current point cloud and obtain the point cloud number of each grid in the current image corresponding to the grid p ( v i ) and signal-to-noise ratio; obtaining point cloud features according to the number of point clouds and the signal-to-noise ratio, wherein the point cloud features include the number of grids whose point cloud number is less than the point cloud threshold n , the number of grids whose signal-to-noise ratio is less than the signal-to-noise ratio threshold m And the overall quality parameters of the point cloud in the corresponding area of ​​the current image H ; The overall quality parameters of the point cloud H The calculation method is as follows: According to the importance of the target area corresponding to the current image, the type of road structure, the frequency of traffic target occurrence, and the possibility of pedestrian intrusion, each grid is assigned a corresponding weight. w ( v i ), 0< w ( v i )<1, the sum of all grid weights in the target area is 1, and the number of point clouds for each grid p ( v i ) and its weight, calculate its conditional entropy: ； The overall quality parameters of the point cloud in the target area are calculated using the following formula: ； in represents the point cloud distribution in the target area, N Represents the sum of all grids in the target area; S5. Based on the environmental features and point cloud features obtained in step S4, determine whether the current location of the sensing base station satisfies the following preset conditions at the same time. If the environmental features and point cloud features meet all the preset conditions at the same time, the site selection procedure ends; otherwise, execute step S6 to adjust the location of the sensing base station; Preset condition 1: There is no obstacle or the occlusion rate of the obstacle is less than the occlusion rate threshold; Preset condition 2: If the lane type is a curve, the lane lines within the preset distance at both ends of the curve fall into the detection area of ​​the camera and lidar; Preset condition 3: The clarity of traffic markings is greater than the preset value; Precondition 4: Point cloud overall quality parameters H greater than the quality parameter threshold; Precondition 5: n ×λ1+ m ×λ2<0.5; n is the number of grids whose point cloud number is less than the point cloud threshold, m is the number of grids with a signal-to-noise ratio less than the signal-to-noise ratio threshold, λ1, λ2 are preset weights, and λ1+ λ2=1 / N , N is the total number of grids; S6. Generate and execute a moving instruction to trigger the movement of the sensing base station. After the sensing base station moves, execute steps S1 to S5 until the site selection procedure is completed.

2. The automatic site selection method according to claim 1, characterized in that: In step S4, the point cloud threshold is ,in, P is the average number of point clouds in a grid.

3. The automatic site selection method according to claim 1, characterized in that: In step S4, the signal-to-noise ratio threshold is set to 70-90.

4. The automatic site selection method according to claim 1, characterized in that: In step S5, the occlusion rate threshold in the preset condition 1 is 10%-25%.

5. The automatic site selection method according to claim 1, characterized in that: In the step S5, the preset distance in the preset condition 2 is 20-50m.

6. The automatic site selection method according to claim 1, characterized in that: In step S5, the preset value in the preset condition 3 is 0.

1.

7. The automatic site selection method according to claim 1, characterized in that: In step S5, the quality parameter threshold in the preset condition 4 is set according to the accuracy of point cloud target recognition required by the site selection.

8. The automatic site selection method according to claim 1, characterized in that: In the step S6, the sensing base station moves according to the judgment result of the preset conditions in step S5. When the preset condition 1 is not met, the sensing base station moves in the direction close to the obstacle; when the preset condition 2 is not met, the sensing base station moves in the direction away from the target area; when the preset condition 3, the preset condition 4 or the preset condition 5 is not met, the horizontal angle and / or the pitch angle of the base station are fine-tuned.

Images