Target vehicle pose detection method and system based on single-line laser radar point cloud

Abstract

The invention provides a target vehicle pose detection method and system based on a single-line laser radar point cloud, and relates to the technical field of vehicle positioning, and the method comprises the following steps: S1, carrying out the preprocessing of point cloud data collected by a laser radar based on a DBSCAN algorithm, screening out noise points, and generating an effective cluster; s2, performing L-shaped feature fitting on each cluster, and calculating a tire center point coordinate; s3, calculating vehicle pose data according to the at least three tire center points, wherein the vehicle pose data comprises a wheelbase, a vehicle center coordinate and a parking angle; and S4, transmitting the vehicle pose data to a parking AGV control system through a communication protocol. The method has the beneficial effects that by adopting the positioning method integrating the laser radar and AGV four-wheel positioning, the vehicle navigation and positioning precision is further improved; a laser radar point cloud is combined with an L-shaped feature fitting algorithm, the pose of a target vehicle is obtained in the walking process of the parking AGV, the pose detection efficiency of the parking AGV is improved, and the walking fluency of the parking AGV is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of vehicle positioning, in particular to a target vehicle pose detection method based on single-line laser radar point cloud and a system thereof. BACKGROUND

[0002] As a core device in the automatic logistics system, the parking AGV plays a very key role in the vehicle parking link and is committed to achieving efficient and accurate parking operation.

[0003] At present, there are various methods for positioning and estimating the pose of the parking AGV for vehicles. Among them, the SLAM based on laser radar is one of the commonly used methods, which scans the surrounding environment through laser radar, constructs a map and determines its own position. In the indoor environment, GPS technology is also used to realize positioning with satellite signals. In addition, the multi-sensor fusion positioning method comprehensively utilizes various sensor information to improve the positioning accuracy. However, the accuracy of these methods is mostly in the range of 5cm-10cm, which still has room for improvement.

[0004] In the factory environment, high-precision positioning AGVs often rely on technologies such as two-dimensional codes, magnetic tracks, and RFID tags. However, this method has strict requirements for the ground conditions, not only does it need to add additional labels as markers to improve positioning accuracy, but it also needs to be regularly maintained to ensure stable system operation.

[0005] Visual navigation is also used in parking AGVs, but it has the problem of light sensitivity. When the environmental light conditions are not good, the pose of the target vehicle cannot be correctly identified in some states, resulting in navigation deviation.

[0006] In general, these existing identification methods have many problems such as slow reaction speed, poor positioning accuracy, limited use scenarios, and high maintenance costs, which need to be improved and optimized to better meet the actual application requirements. SUMMARY

[0007] The present application overcomes the shortcomings in the prior art and provides a target vehicle pose detection method based on single-line laser radar point cloud and a system thereof, which has the advantages of high precision and fast reaction.

[0008] To solve the above technical problems, the present application is realized by the following technical scheme:

[0009] The target vehicle pose detection method based on single-line laser radar point cloud comprises the following steps:

[0010] S1. Preprocess the point cloud data collected by the laser radar based on the DBSCAN algorithm, screen out noise points and generate effective clustering clusters;

[0011] S2. Perform L-type feature fitting on each cluster, calculate tire center point coordinates;

[0012] S3. Calculate vehicle pose data including wheelbase, vehicle center coordinates and parking angle according to at least three tire center points;

[0013] S4. Transmit vehicle pose data to parking AGV control system through communication protocol.

[0014] Further, the S1 comprises:

[0015] S11. According to the preset parking safety line boundary, mark the point cloud data beyond the safety line or less than 50 cm from the radar in the radar coordinate system as noise points;

[0016] S12. Calculate the angle Φ between the diagonal coordinates (x, y) of the safety line and the X axis, and divide the angle interval based on Φ to dynamically limit the maximum detection distance of the point cloud;

[0017] S13. Perform DBSCAN clustering on the point cloud data, and group the point set with a Euclidean distance less than a threshold value eps and a number of neighboring points greater than minPts into the same cluster.

[0018] Further, the angle interval division and distance calculation method in S12 are:

[0019] (1) Interval [-90°, -90°+Φ*(180 / π))

[0020]

[0021] (2) Interval [-90°+Φ*(180 / π), 90°-Φ*(180 / π))

[0022]

[0023] (3) Interval [90°-Φ*(180 / π), 90°]

[0024]

[0025] Where x and y are the diagonal coordinate values of the safety line.

[0026] The Euclidean distance calculation formula in S13 is

[0027]

[0028] Where (x, y) and (xi, yi) are the coordinates of two points.

[0029] Further, the S2 comprises:

[0030] S21. Extract the closest point to the radar in the cluster and divide the cluster point set into two subsets;

[0031] S22. Use the least squares method to fit the linear equations of the two subsets respectively;

[0032] S23. When the absolute difference between the angle between two straight lines and 90° is less than the threshold, it is determined to be a tire point cloud, and the coordinates of the corner point and the coordinates of the endpoint far from the corner point are calculated.

[0033] S24. Calculate the coordinates of the tire center point based on the coordinates of the two endpoints.

[0034] Furthermore, the linear fitting in S22 includes:

[0035] (1) Calculate the statistic

[0036]

[0037] (2) Calculate the mean

[0038]

[0039] (3) Calculate the slope and intercept.

[0040]

[0041] (4) Convert to the general form ax + by + c = 0

[0042]

[0043] (5) Normalization

[0044]

[0045] (6) Determine the endpoints of the line segment

[0046]

[0047] Furthermore, the calculation of the corner coordinates and the coordinates of the endpoints far from the corner in S23 includes: (1) calculating the angle between the two lines.

[0048] θ = arccos(|a1a2+b1b2|)

[0049] (2) Calculate corner points

[0050]

[0051] (3) Find the endpoints of the two lines that are farthest from the corner point, and the center point is...

[0052]

[0053] Further, the S3 comprises:

[0054] S31. When more than 3 tire center points are identified, calculate the distance between each center point

[0055]

[0056] S32. Take the middle value in the distance as the wheelbase

[0057] wheelbase=d mid

[0058] S33. Determine the vehicle center coordinates according to the maximum distance endpoints:

[0059]

[0060] S34. Calculate the parking angle offset according to the vector of the side where the middle value distance is located.

[0061]

[0062] Further, the communication protocol in the S4 adopts the Modbus TCP protocol.

[0063] The application also claims a parking AGV positioning system, comprising:

[0064] The laser radar with an installation height lower than a preset height has a scanning range limited in [-90°, 90°];

[0065] The processor is used for executing the pose detection method in any one of claims 1 to 8;

[0066] The communication module is used for transmitting the pose data to the AGV control system.

[0067] Further, the preset height is 50 cm.

[0068] Compared with the prior art, the application has the beneficial effects that:

[0069] 1. The positioning method adopting the laser radar and the AGV four-wheel positioning fusion further improves the precision of vehicle navigation positioning;

[0070] 2. The laser radar point cloud combined with the L-shaped feature fitting algorithm is used to obtain the target vehicle pose in the parking AGV walking process, so as to improve the parking AGV pose detection efficiency and improve the parking AGV walking fluency. BRIEF DESCRIPTION OF DRAWINGS

[0071] The accompanying drawings are used to provide a further understanding of the application, together with the embodiments of the application, to explain the application, and do not constitute a limitation to the application, and in the drawings:

[0072] Figure 1 is a flowchart of the target vehicle pose detection method based on single-line laser radar point cloud of the present application. DETAILED DESCRIPTION

[0073] The preferred embodiments of the present application are described below in conjunction with the accompanying drawings, it should be understood that the preferred embodiments described herein are only used to illustrate and explain the present application, and are not used to limit the present application.

[0074] As shown in Figure 1 The present application claims a target vehicle pose detection method based on single-line laser radar point cloud: comprising the following steps:

[0075] S1. Preprocess the point cloud data collected by the laser radar based on the DBSCAN algorithm, screen out noise points and generate effective clustering clusters;

[0076] S2. Perform L-type feature fitting on each clustering cluster and calculate the tire center point coordinates;

[0077] S3. Calculate the vehicle pose data including wheelbase, vehicle center coordinates and parking angle according to at least three tire center points;

[0078] S4. Transmit the vehicle pose data to the parking AGV control system through the communication protocol.

[0079] Wherein, S1 includes:

[0080] S11. According to the preset parking safety line boundary, mark the point cloud data beyond the safety line or less than 50cm from the radar in the radar coordinate system as noise points;

[0081] That is, first, the position of the preset parking safety line boundary in the radar coordinate system is determined. The boundary range of the safety line is determined by the pre-set rule, and the point cloud data beyond the safety line boundary is directly marked as noise points.

[0082] At the same time, for the point cloud data less than 50cm from the radar, it is also marked as noise points due to the influence of ground reflection and other factors. In this way, some point clouds that do not meet the target vehicle detection range can be removed initially.

[0083] S12. Calculate the angle Φ between the diagonal coordinates (x, y) of the safety line and the X axis, and divide the angle interval based on Φ to dynamically limit the maximum detection distance of the point cloud;

[0084] Wherein, the angle interval division and distance calculation method in S12 is:

[0085] (1) Interval [-90°, -90°+Φ*(180 / π))

[0086]

[0087] (2) interval [-90°+Φ*(180 / π), 90°-Φ*(180 / π))

[0088]

[0089] (3) interval [90°-Φ*(180 / π), 90°]

[0090]

[0091] wherein x, y are the diagonal coordinate values of the security line.

[0092] The maximum detection distance of the point cloud is dynamically limited according to different angle intervals. For example, different distance upper limit values are set in different intervals, so that the detection range of the point cloud can be reasonably limited according to different angle positions that the vehicle may appear, and unnecessary calculation amount is reduced.

[0093] S13. Perform DBSCAN clustering on the point cloud data, and group the point set with a Euclidean distance less than a threshold value eps and a neighbor point number greater than minPts into the same clustering cluster.

[0094] The DBSCAN algorithm is used for clustering operation. For each point in the point cloud data, the Euclidean distance between the point and other points is calculated.

[0095]

[0096] wherein (x, y) and (x i , y i ) are the coordinates of two points.

[0097] By traversing the point cloud data, the neighbor points with a distance value less than the threshold value eps are obtained. If the number of neighbor points is higher than the threshold value minPtr, the point cloud data and the corresponding neighbor points are marked as the same clustering cluster, otherwise they are marked as noise points. Thus, the point cloud data preprocessing process ends.

[0098] Further, the S2 comprises:

[0099] S21. Extracting the point closest to the radar in the clustering cluster, and dividing the cluster point set into two subsets;

[0100] In each clustering cluster, the point closest to the radar is found by comparing the distance of each point to the radar. The other points in the clustering cluster are divided into two subsets according to certain rules based on the closest point. For example, the points can be divided according to the relative position relationship in the clustering cluster, such as the angle range.

[0101] S22. Fitting straight line equations for the two subsets respectively using least square method;

[0102] The straight line fitting in S22 includes:

[0103] (1) Calculate statistics

[0104]

[0105] (2) Calculate mean

[0106]

[0107] (3) Calculate slope and intercept

[0108]

[0109] (4) Convert to general form ax + by + c = 0

[0110]

[0111] (5) Normalize

[0112]

[0113] (6) Determine line segment endpoints

[0114]

[0115] S23. When the absolute difference between the angle between the two straight lines and 90° is less than a threshold value, determine that it is a tire point cloud, calculate the coordinates of the corner point and the coordinates of the endpoint away from the corner point;

[0116] S24. Calculate the coordinates of the tire center point according to the two endpoint coordinates.

[0117] Further, the calculation of the corner point coordinates and the endpoint coordinates away from the corner point in S23 includes:

[0118] (1) Calculate the angle between the two straight lines

[0119] θ = arccos (|a1a2 + b1b2|)

[0120] (2) Calculate the corner point

[0121]

[0122] (3) Find the endpoints of the two straight lines farthest from the corner point, and the center point is

[0123]

[0124] Further, the S3 includes:

[0125] S31. When more than 3 tire center points are identified, calculate the distance between each center point

[0126]

[0127] S32. Take the middle value in the distance as the wheelbase

[0128] wheelbase = d mid

[0129] S33. Determine the vehicle center coordinates according to the maximum distance endpoints:

[0130]

[0131] S34. Calculate the parking angle offset according to the vector of the edge where the middle value distance is located.

[0132]

[0133] In this embodiment, the communication protocol in S4 adopts the Modbus TCP protocol.

[0134] The application also claims a parking AGV positioning system, comprising: a laser radar with an installation height lower than a preset height, a processor, and a communication module; in this embodiment, the preset height is 50 cm; due to the complexity of the parking environment, the laser radar point cloud data needs to limit the measurement range within the demarcated parking safety line, by calculating the angle between the diagonal coordinates of the parking safety line in the radar coordinate system and the X axis of the coordinate system, the radar measurement data exceeding the safety line and the part of data with a distance lower than 50 cm from the radar are framed onto the safety line, and this part of data is marked as noise points in advance, so as to avoid affecting the clustering data L-shaped feature fitting calculation when the vehicle is parked near the safety line. Therefore, the installation height is lower than 50 cm, and such design can ensure that the laser radar effectively scans the relevant information of the target vehicle, while avoiding unnecessary interference due to being too high.

[0135] The scanning range of the laser radar is limited to [-90°, 90°], the maximum distance of the point cloud data is limited so that it does not exceed the limited rectangular boundary, and the area where the target vehicle may appear can be fully covered, providing a complete information basis for subsequent data collection

[0136] The processor is used to perform the above-mentioned target vehicle pose detection method based on single-line laser radar point cloud.

[0137] After the pose data required by the parking AGV is calculated, the communication module transmits the calculated data to the parking AGV through the modbus TCP protocol, and the auxiliary pose detection method part is completed.

[0138] The application adopts the method of laser radar and AGV four-wheel positioning, improves the precision of vehicle navigation positioning, and accurately perceives the position and attitude of the vehicle in the parking lot by organically combining the accurate environment information obtained by the laser radar with the AGV four-wheel positioning technology, effectively reduces the positioning error, and provides a strong guarantee for the precise navigation of the vehicle.

[0139] The single-line laser radar point cloud is adopted, and the L-shaped feature fitting algorithm is combined to obtain the pose of the target vehicle in the walking process of the parking AGV, so that the efficiency of the pose detection of the parking AGV is improved, accurate pose data can be obtained in a short time, and the fluency of the walking of the parking AGV is improved.

[0140] Finally, it should be noted that: the above is only the preferred embodiment of the present application, and is not used to limit the present application, although the present application is described in detail with reference to the embodiments, for those skilled in the art, the technical solutions recorded in the foregoing embodiments can be modified, or some technical features can be replaced, but any modification, equivalent replacement, improvement, etc. within the spirit and principles of the present application should be included in the protection scope of the present application.

Claims

1. A target vehicle pose detection method based on single-line laser radar point cloud, characterized in that, The method comprises the following steps: S1. Preprocessing the point cloud data collected by the laser radar based on the DBSCAN algorithm, screening out noise points and generating effective cluster; S2. Performing L-shaped feature fitting on each cluster, calculating the tire center point coordinates; S3. Calculating the vehicle pose data based on at least three tire center points, including wheelbase, vehicle center coordinates and parking angle; S4. Transmitting the vehicle pose data to the parking AGV control system through the communication protocol.

2. The single-line lidar point cloud based target vehicle pose detection method of claim 1, wherein, The S1 comprises: S11. According to the preset parking safety line boundary, mark the point cloud data beyond the safety line or less than 50cm from the radar in the radar coordinate system as noise points; S12. Calculate the angle Φ between the diagonal coordinates (x, y) of the safety line and the X axis, and divide the angle interval based on Φ to dynamically limit the maximum detection distance of the point cloud; S13. Perform DBSCAN clustering on the point cloud data, and group the point set with a Euclidean distance less than the threshold value eps and a neighbor point number greater than minPts into the same cluster.

3. The single-line lidar point cloud based target vehicle pose detection method of claim 1, wherein, The angle interval division and distance calculation method in S12 is: (1) Interval [-90°, -90°+Φ*(180 / π)) (2) Interval [-90°+Φ*(180 / π), 90°-Φ*(180 / π)) (3) Interval [90°-Φ*(180 / π), 90°] Where x and y are the diagonal coordinate values of the safety line; The Euclidean distance calculation formula in S13 is where (x, y) and (x i , y i ) are two point coordinates.

4. The single-line lidar point cloud based target vehicle pose detection method of claim 1, wherein, The S2 comprises: S21. Extract the point closest to the radar in the cluster, and divide the cluster point set into two subsets; S22. Fit the straight line equations of the two subsets respectively using the least squares method; S23. When the absolute difference between the included angle of the two straight lines and 90° is less than the threshold value, it is determined as a tire point cloud, and the corner point coordinates and the endpoint coordinates away from the corner point are calculated; S24. Calculate the tire center point coordinates according to the two endpoint coordinates.

5. The single-line lidar point cloud based target vehicle pose detection method of claim 4, wherein, The straight line fitting in S22 comprises: (1) Calculate the statistic (2) Calculate the mean (3) Calculate the slope slope and intercept intercept (4) Convert to general formula ax+by+c=0 (5) Normalize (6) Determine the line segment endpoints 6. The single-line lidar point cloud based target vehicle pose detection method of claim 4, wherein, The calculation of the corner point coordinates and the endpoint coordinates away from the corner point in S23 comprises: (1) Calculate the included angle of the two straight lines θ=arccos(|a1a2+b1b2|) (2) Calculate the corner point (3) Find the endpoints of the two straight lines farthest from the corner point, and the center point is 7. The single-line lidar point cloud based target vehicle pose detection method of claim 1, wherein, The S3 comprises: S31. When more than three tire center points are identified, calculate the distance between each center point S32. Take the middle value in the distance as the wheelbase wheelbase = d mid S33. Determine the vehicle center coordinates according to the maximum distance endpoint: S34. Calculate the parking angle offset according to the vector of the side where the middle value distance is located.

8. The single-line lidar point cloud based target vehicle pose detection method of claim 1, wherein, The communication protocol in S4 uses Modbus TCP protocol.

9. A parking AGV positioning system, characterized in that, Comprise: The laser radar with an installation height lower than the preset height, and the scanning range is limited to [-90°, 90°]; The processor is used to execute the pose detection method of any one of claims 1 to 8; The communication module is used to transmit the pose data to the AGV control system.

10. The parking AGV positioning system of claim 9, wherein, The preset height is 50cm.

Images