A 3D laser radar-based obstacle detection method for unmanned mine car

Abstract

The present application relates to a kind of unmanned mine car obstacle detection method based on 3D laser radar, belong to obstacle detection technical field, solve the over detection, the problem such as the rate of missed detection in the existing unmanned mine car obstacle detection process exists.Processing includes: preset multiple path sampling points on the fixed driving track of mining area operation;Periodically collect laser point cloud data and the current position of unmanned mine car during the driving process of unmanned mine car along driving track;Each periodically collected group of laser point cloud data and the current position of unmanned mine car, all use the following way to carry out obstacle detection: the first path sampling point after unmanned mine car from current position along driving track travels predetermined distance is as predetermined position, based on the north angle deviation and distance between predetermined position and path sampling point in predetermined position travel direction, determine region of interest;Obstacle detection is carried out based on the laser point cloud data of region of interest.

Description

Technical Field

[0001] This invention relates to the field of obstacle detection technology, and in particular to an obstacle detection method for unmanned mining vehicles based on 3D LiDAR. Background Technology

[0002] Open-pit mines present complex road environments, often including obstacles such as earthen slopes, deep pits, retaining walls, pedestrians, and other auxiliary vehicles, which places high demands on the safety of unmanned mining truck operations. In recent years, with the development of information technology and intelligent systems in mining areas, more intelligent unmanned mining truck systems are being increasingly applied to production operations. Among these systems, the environmental intelligent perception system, a crucial component of unmanned mining trucks, primarily ensures the safe and reliable operation of the trucks in the mining environment. It is responsible for effectively detecting pedestrians, vehicles, and other ground obstacles, and promptly transmitting obstacle information to the decision-making system.

[0003] For obstacle detection on mining roads, region of interest extraction is a crucial step. It involves cropping the original point cloud, retaining only the point cloud data along the driving path, which effectively improves computational efficiency and reduces false negatives. Traditional methods use curb detection to crop the obtained point cloud, but in mining areas, flat roads without curbs are common, leading to an excessive number of drivable surfaces and over-detection, which is detrimental to improving computational efficiency. Another method is point cloud segmentation based on the Z-axis coordinates of laser point clouds, requiring detailed classification of obstacle heights, which can easily lead to false positives. Regarding obstacle detection algorithms, several methods have been proposed, such as grid map methods that compress 3D data into a 2D plane and perform clustering, and feature image methods that capture point clouds frame by frame and use visual methods for detection. However, these methods only obtain target information on a single 2D plane and often fail to effectively describe objects in practical applications. Another type of detection method aggregates scattered data points into independent categories based on differences generated by some pre-determined criteria (such as density, distance, or hierarchy). However, the clustering radius of this type of clustering algorithm is not adjustable, and it does not take into account the impact of the distance between obstacles and unmanned mining vehicles on the clustering process, resulting in a high rate of missed detections in obstacle detection results. Summary of the Invention

[0004] Based on the above analysis, the present invention aims to provide an obstacle detection method for unmanned mining trucks based on 3D LiDAR, in order to solve the problems of over-detection and high false negative rate in the existing obstacle detection process of unmanned mining trucks.

[0005] This invention discloses an obstacle detection method for unmanned mining trucks based on 3D LiDAR, comprising:

[0006] Multiple path sampling points are pre-set on the fixed driving trajectory of the mining area; as the unmanned mining vehicle travels along the driving trajectory, laser point cloud data and the current position of the unmanned mining vehicle are periodically collected;

[0007] Each periodically collected set of laser point cloud data and the current position of the unmanned mining vehicle are used for obstacle detection in the following manner:

[0008] The first path sampling point after the unmanned mining vehicle travels a predetermined distance from its current position along the driving trajectory is taken as the predetermined position. Based on the predetermined position and the northward deviation and distance between the predetermined position and the path sampling points in the direction of travel of the predetermined position, the region of interest is determined.

[0009] Obstacle detection is performed based on laser point cloud data of the region of interest.

[0010] Among them, the laser point cloud data is collected using 3D LiDAR.

[0011] Based on the above solution, the present invention also makes the following improvements:

[0012] Furthermore, the location of the path sampling point and the current location of the unmanned mining vehicle both include: the coordinates X on the x-axis and Y on the y-axis in the world coordinate system, as well as the north angle Theta.

[0013] Furthermore, the region of interest is determined by performing the following operations:

[0014] The path sampling points are sorted in order from near to far from the predetermined position in the direction of travel to form a path sampling point sequence;

[0015] The predetermined location is taken as the starting point P of the j-th region of interest. start (j), at this time, j = 1;

[0016] Extract path sampling points sequentially from the path sampling point sequence. When extracting the i-th path sampling point, perform a relative relationship judgment:

[0017] Get P start (j) The north angle deviation θ(i,j) and straight-line distance l(i,j) from the i-th path sampling point.

[0018] like Then the i-th path sampling point is taken as the termination point P of the j-th region of interest. stop (j), iteration ends;

[0019] like And θ(i,j)≤θ set Update i to i+1, extract the i-th path sampling point, and repeat the relative relationship judgment;

[0020] like And θ(i,j)>θ set Then the i-th path sampling point is taken as the termination point P of the j-th region of interest. stop (j); with the termination point P of the j-th region of interest stop (j) is the starting point P of the (j+1)th region of interest. start (j+1), establish the (j+1)th region of interest, update j to j+1; update i to i+1, extract the i-th path sampling point, and repeat the relative relationship judgment;

[0021] After the iteration is completed, the starting point and ending point of one or more regions of interest are obtained. Based on the obtained starting point and ending point, the region of interest is determined.

[0022] Where, θ set Indicates the north angle deviation threshold, l set L(0) = 0, where L(k) represents the cumulative distance threshold. When k ≥ 1, L(k) represents the straight-line distance between the starting point and the ending point of the kth region of interest. i = 1, 2, ..., N, where N is the length of the path sampling point sequence.

[0023] Furthermore, based on the obtained start and end points, the region of interest is determined, including:

[0024] Expand the starting and ending points of each region of interest by the road width to obtain the corner points on both sides of the starting and ending points. Then, connect the two corner points on the same side of the starting and ending points of each region of interest, the corner points on both sides of the starting point, the two corner points on the other side of the starting and ending points, and the corner points on both sides of the ending point in sequence to obtain the complete region of interest.

[0025] Multiple complete regions of interest are integrated and identified as the region of interest.

[0026] Furthermore, the cumulative distance threshold is determined based on the predetermined distance, the driving speed of the unmanned mining vehicle, and the detection range of the 3D lidar.

[0027] Furthermore, the cumulative distance threshold is 30m; the northward angle deviation threshold is 5°.

[0028] Furthermore, multiple path sampling points are evenly pre-placed on the fixed driving trajectory, with a distance of 0.2m between two adjacent path sampling points.

[0029] Furthermore, obstacle detection is performed based on the laser point cloud data of the region of interest, including:

[0030] Process the laser point cloud data of the region of interest to obtain the non-ground point cloud data of the region of interest;

[0031] Obstacle detection is performed on non-ground point cloud data within the region of interest.

[0032] Furthermore, non-terrestrial point cloud data within the region of interest is obtained through the following methods:

[0033] The laser point cloud data of the region of interest is filtered based on the voxel filtering method;

[0034] The beam-based ground segmentation method performs ground segmentation on filtered laser point cloud data to obtain non-ground point cloud data within the region of interest.

[0035] Furthermore, obstacle detection is performed on the non-ground point cloud data within the region of interest by performing the following operations:

[0036] Step S1: Randomly select a non-ground point cloud data point, label it as a category, and obtain the cluster radius corresponding to the non-ground point cloud data point according to formula (1).

[0037]

[0038] In the formula, λ is the horizontal angular resolution of the 3D LiDAR, λ is a variable threshold parameter, and r is the distance between the non-ground point cloud data and the center point of the 3D LiDAR.

[0039] The non-ground point cloud data is used as the cluster center, and the corresponding cluster radius is used as the cluster radius corresponding to the cluster center.

[0040] Step S2: Determine whether there is data that meets the clustering requirements among all unclassified non-ground point cloud data. If so, assign the non-ground point cloud data that meets the clustering requirements to the category to which the cluster center belongs, and use each newly assigned non-ground point cloud data as the updated cluster center, and obtain the cluster radius corresponding to the updated cluster center. Repeat step S2 until there is no more non-ground point cloud data that meets the clustering requirements; divide all non-ground point cloud data that meets the clustering requirements into the target of the category.

[0041] Step S3: Determine whether there is unclassified non-ground point cloud data. If not, end; if so, proceed to step S4.

[0042] Step S4: Select any unclassified non-ground point cloud data, label it with a new category, and use the unclassified non-ground point cloud data as the updated cluster center. Obtain the cluster radius corresponding to the updated cluster center, and then execute step S2; until all non-ground point cloud data are divided into a certain category of target.

[0043] The clustering requirement refers to the following: the Euclidean distance between unclassified non-ground point cloud data and the cluster center is less than the cluster radius corresponding to that cluster center;

[0044] Step S5: Determine the number of non-ground point cloud data contained in each category of target, and detect targets of categories that meet the number threshold as obstacles.

[0045] Compared with the prior art, the present invention can achieve at least one of the following beneficial effects:

[0046] This invention discloses a method for obstacle detection in unmanned mining trucks based on 3D LiDAR. This method dynamically determines the region of interest (ROI) based on path sampling points along the fixed driving trajectory in the mining area and the current position of the unmanned mining truck, and then performs obstacle detection on the LiDAR point cloud data within the dynamically determined ROI. This method effectively solves the problems of missing curbs due to irregular road surfaces in mining areas and over-detection of obstacles due to excessively wide road surfaces.

[0047] Meanwhile, during obstacle detection, the clustering radius can be dynamically updated based on the distance between the non-ground point cloud data and the center point of the 3D LiDAR, thus effectively meeting the detection requirements of obstacles at different distances from the unmanned mining vehicle and reducing the false negative rate.

[0048] In this invention, the above-described technical solutions can be combined with each other to achieve more preferred combinations. Other features and advantages of this invention will be set forth in the following description, and some advantages may become apparent from the description or be learned by practicing the invention. The objects and other advantages of this invention can be realized and obtained from what is particularly pointed out in the description and drawings. Attached Figure Description

[0049] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts.

[0050] Figure 1 This is a flowchart of the obstacle detection method for unmanned mining trucks based on 3D LiDAR in an embodiment of the present invention;

[0051] Figure 2 This is an example of a fixed driving trajectory in an embodiment of the present invention;

[0052] Figure 3 (a) and (b) are laser point cloud data distribution diagrams before and after the determination of the region of interest corresponding to the straight-moving vehicle trajectory in the embodiment of the present invention;

[0053] Figure 4 (a) and (b) are laser point cloud data distribution diagrams before and after the determination of the region of interest corresponding to the curved driving trajectory in the embodiment of the present invention;

[0054] Figure 5 This is an example of the original mining area scene in an embodiment of the present invention;

[0055] Figure 6 (a) and (b) are point cloud effect diagrams before and after ground segmentation in the embodiments of the present invention, respectively. Detailed Implementation

[0056] Preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, which form part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not intended to limit the scope of the present invention.

[0057] A specific embodiment of the present invention discloses an obstacle detection method for unmanned mining trucks based on 3D LiDAR, the flowchart of which is shown below. Figure 1 As shown, it includes the following steps:

[0058] Step S1: Pre-set multiple path sampling points on the fixed driving trajectory of the mining operation; during the unmanned mining vehicle's journey along the driving trajectory, periodically collect laser point cloud data and the current position of the unmanned mining vehicle; wherein, the laser point cloud data is collected using 3D LiDAR;

[0059] In step S1, the laser point cloud data is acquired using a 3D LiDAR. The horizontal field of view of the LiDAR is 360°, and the vertical field of view is 45° (±22.5°). During use, a lot of redundant point cloud data is generated, which significantly reduces the calculation speed and affects the real-time performance of the entire method. Since only obstacles within a certain range centered on the vehicle's trajectory will affect the unmanned mining vehicle's movement during mining operations, it is only necessary to determine whether obstacles exist within that range based on the point cloud data. This embodiment proposes a scheme to dynamically extract the region of interest from the acquired point cloud and dynamically retain the point cloud data within the drivable area; the specific process is shown in step S2:

[0060] Step S2: Periodically collect a set of laser point cloud data and the current position of the unmanned mining vehicle, and perform obstacle detection using the following method:

[0061] Step S21: Take the first path sampling point after the unmanned mining vehicle has traveled a predetermined distance from its current position along the driving trajectory as the predetermined position. Based on the predetermined position and the northward deviation and distance between the path sampling points in the direction of travel of the predetermined position, determine the region of interest.

[0062] It should be noted that obstacle detection for unmanned mining vehicles must be predictive, meaning it should be able to identify obstacles along a future section of the driving path in advance, allowing the unmanned mining vehicle to react accordingly. Therefore, this embodiment selects the first path sampling point after the unmanned mining vehicle has traveled a predetermined distance along the driving trajectory from its current position as the predetermined position. The region of interest is determined based on the northward deviation and distance between the predetermined position and path sampling points along the direction of travel from the predetermined position. In practical applications, a predetermined distance of 4 meters is preferred.

[0063] Preferably, in this embodiment, the location of the path sampling point and the current location of the unmanned mining vehicle both include: the coordinates X on the x-axis and Y on the y-axis in the world coordinate system, as well as the north angle Theta.

[0064] To ensure the accuracy of the region of interest determination, in this embodiment, multiple path sampling points are pre-set evenly on a fixed driving trajectory, with a distance of 0.2m between two adjacent path sampling points.

[0065] In step S21, the region of interest is determined in the following way:

[0066] The path sampling points are sorted in order from near to far from the predetermined position in the direction of travel to form a path sampling point sequence;

[0067] The predetermined location is taken as the starting point P of the j-th region of interest. start (j), at this time, j = 1;

[0068] Extract path sampling points sequentially from the path sampling point sequence. When extracting the i-th path sampling point, perform a relative relationship judgment:

[0069] Get P start (j) The north angle deviation θ(i,j) and straight-line distance l(i,j) from the i-th path sampling point.

[0070] like Then the i-th path sampling point is taken as the termination point P of the j-th region of interest. stop (j), iteration ends;

[0071] like And θ(i,j)≤θ set Update i to i+1, extract the i-th path sampling point, and repeat the relative relationship judgment;

[0072] like And θ(i,j)>θ set Then the i-th path sampling point is taken as the termination point P of the j-th region of interest. stop (j); with the termination point P of the j-th region of interest stop(j) is the starting point P of the (j+1)th region of interest. start (j+1), establish the (j+1)th region of interest, update j to j+1; update i to i+1, extract the i-th path sampling point, and repeat the relative relationship judgment;

[0073] After the iteration is completed, the starting point and ending point of one or more regions of interest are obtained. Based on the obtained starting point and ending point, the region of interest is determined.

[0074] Where, θ set Indicates the north angle deviation threshold, l set L(0) = 0, where L(k) represents the cumulative distance threshold. When k ≥ 1, L(k) represents the straight-line distance between the starting point and the ending point of the kth region of interest. i = 1, 2, ..., N, where N is the length of the path sampling point sequence.

[0075] Preferably, the starting and ending points of each region of interest are expanded by the road width to obtain the corner points on both sides of the starting and ending points. Then, the two corner points on the same side of the starting and ending points of each region of interest, the corner points on both sides of the starting point, the two corner points on the other side of the starting and ending points, and the corner points on both sides of the ending point are connected sequentially to obtain a complete region of interest. Multiple complete regions of interest are integrated to determine the region of interest. In the actual implementation of this scheme, a road width of 4m is preferred.

[0076] Preferably, in this embodiment, the cumulative distance threshold is determined based on the predetermined distance, the speed of the unmanned mining vehicle, and the detection range of the 3D lidar. Preferably, the cumulative distance threshold is 30m; the northward deviation threshold is 5°.

[0077] Figure 3 (a) and 3(b) are laser point cloud data distribution maps before and after the determination of the region of interest corresponding to the straight-moving vehicle trajectory; Figure 4 (a) and (b) are the distribution maps of laser point cloud data before and after the determination of the region of interest corresponding to the curve driving trajectory. The analysis shows that by determining the region of interest, the amount of laser point cloud data can be greatly reduced, thereby effectively reducing the amount of laser point cloud data that needs to be processed in subsequent work and improving work efficiency.

[0078] Step S22: Obstacle detection is performed based on the laser point cloud data of the region of interest. Specifically,

[0079] Step S221: Process the laser point cloud data of the region of interest to obtain the non-ground point cloud data of the region of interest;

[0080] The laser point cloud data of the region of interest is filtered based on the voxel filtering method;

[0081] The beam-based ground segmentation method performs ground segmentation on filtered laser point cloud data to obtain non-ground point cloud data within the region of interest.

[0082] Specifically, voxel filtering is a commonly used filtering method for 3D point cloud data. It can reduce the number of points in the computation while preserving the characteristics of each part of the point cloud. Assuming the point cloud space is within a cuboid, the point cloud space is divided into cuboid units of fixed size, with dimensions L, H, and D along the X, Y, and Z axes, respectively.

[0083] The point cloud data within the region of interest contains not only obstacle points but also a large number of ground points. However, a large number of ground points can interfere with obstacle detection, so it is necessary to remove them. In this embodiment, a ground segmentation method based on wires is used to segment the ground. This algorithm projects the three-dimensional point cloud onto a two-dimensional plane. First, the point cloud is classified according to the angle. The distance between the point cloud data at the same angle and the lidar in the horizontal direction is calculated according to formula (1), and then sorted in order of increasing distance.

[0084]

[0085] x and y represent the coordinate values ​​on the x and y axes of the point cloud data, respectively;

[0086] After obtaining the sorting of point cloud data at each angle, for each point cloud data, it is determined whether the slope between the point cloud data and the nearest point cloud is less than or equal to the slope threshold. If so, the point cloud data is determined to be a ground point; if it is greater, it is further determined whether the slope between the point cloud data and the center point of the lidar exceeds the road surface threshold. If it exceeds the threshold, the point cloud data is determined to be a ground point; otherwise, it is determined to be a non-ground point.

[0087] The slope threshold and road surface threshold can be set adaptively based on the overall smoothness of the road surface in the mining area;

[0088] The formula for calculating the slope between two point cloud datasets is as follows:

[0089]

[0090] In the formula, z i+1 -z i R represents the height difference between two adjacent points. i+1 -R i This represents the distance difference between two adjacent points. The original scene is as follows: Figure 5 As shown, the point cloud renderings before and after ground segmentation are as follows: Figure 6 As shown in (a) and 6(b).

[0091] Step S222: Perform obstacle detection on the non-ground point cloud data within the region of interest.

[0092] Obstacle detection is a core component of unmanned driving in mining areas. The key challenges in obstacle detection lie in selecting the clustering radius and determining the obstacle criteria. To more effectively identify obstacles, the number of laser point cloud data points distributed across different obstacles varies depending on their location: for distant obstacles, the number of point cloud points is relatively small; for closer obstacles, the number of point cloud points is large. Therefore, this paper uses a method based on a variable clustering radius to detect obstacles within the region of interest. Specifically:

[0093] Step S2221: Randomly select a non-ground point cloud data point, label it as a category, and obtain the cluster radius corresponding to the non-ground point cloud data point according to formula (1).

[0094]

[0095] In the formula, λ is the horizontal angular resolution of the 3D LiDAR, λ is a variable threshold parameter, and r is the distance between the non-ground point cloud data and the center point of the 3D LiDAR.

[0096] The non-ground point cloud data is used as the cluster center, and the corresponding cluster radius is used as the cluster radius corresponding to the cluster center.

[0097] Step S2222: Determine whether there is data that meets the clustering requirements among all unclassified non-ground point cloud data. If so, assign the non-ground point cloud data that meets the clustering requirements to the category to which the cluster center belongs, and use each newly assigned non-ground point cloud data as the updated cluster center, and obtain the cluster radius corresponding to the updated cluster center. Repeat step S2222 until there is no more non-ground point cloud data that meets the clustering requirements; divide all non-ground point cloud data that meets the clustering requirements into the target of the category.

[0098] Step S2223: Determine whether there is unclassified non-ground point cloud data. If not, end; if so, proceed to step S2224.

[0099] Step S2224: Select any unclassified non-ground point cloud data, label it with a new category, and use the unclassified non-ground point cloud data as the updated cluster center. Obtain the cluster radius corresponding to the updated cluster center, and then execute step S2222; until all non-ground point cloud data are divided into a certain category of target.

[0100] The clustering requirement refers to the following: the Euclidean distance between unclassified non-ground point cloud data and the cluster center is less than the cluster radius corresponding to that cluster center;

[0101] Step S2225: Determine the number of non-ground point cloud data contained in each category of target, and detect the targets of the category that meet the number threshold as obstacles.

[0102] If a certain category of targets contains relatively little non-ground point cloud data, it indicates that the target in that category may be noise or clutter; only when a certain category of targets contains a large amount of non-ground point cloud data can it be considered an obstacle. By performing the above process, one or more obstacles can be identified. After repeated experiments, a threshold of 4 is preferred. At this threshold, it is possible to accurately determine whether each category of targets is an obstacle, thus completing the obstacle detection process.

[0103] During obstacle detection, for each cluster center, the cluster radius corresponding to the cluster center is dynamically adjusted according to formula (4). This can effectively improve the problem of inconsistent cluster radii caused by the distance of obstacles, thereby more effectively identifying obstacles and reducing the false detection rate.

[0104] In summary, the obstacle detection method for unmanned mining trucks based on 3D LiDAR provided in this embodiment can dynamically determine the region of interest (ROI) based on path sampling points on the fixed driving trajectory of the mining operation and the current position of the unmanned mining truck, and then perform obstacle detection on the LiDAR point cloud data within the dynamically determined ROI. This method effectively solves the problems of missing curbs due to irregular road surfaces in mining areas and over-detection of obstacles due to excessively wide road surfaces. Furthermore, during obstacle detection, the clustering radius can be dynamically updated based on the distance between the non-ground point cloud data and the center point of the 3D LiDAR, thus well satisfying the detection process for obstacles at different distances from the unmanned mining truck and effectively reducing the false negative rate.

[0105] Those skilled in the art will understand that all or part of the processes of the methods described in the above embodiments can be implemented by a computer program instructing related hardware, and the program can be stored in a computer-readable storage medium. The computer-readable storage medium may be a disk, optical disk, read-only memory, or random access memory, etc.

[0106] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A 3D laser radar-based obstacle detection method for unmanned mine vehicles, characterized by, The method comprises the following steps: a plurality of path sampling points are preset on a fixed driving track in a mining area; laser point cloud data and the current position of the unmanned mine vehicle are periodically collected during the driving of the unmanned mine vehicle along the driving track; the following method is used for obstacle detection for each set of periodically collected laser point cloud data and the current position of the unmanned mine vehicle: a first path sampling point after the unmanned mine vehicle travels a predetermined distance from the current position along the driving track is taken as a predetermined position, a region of interest is determined based on the deviation of the north angle and the distance between the predetermined position and the path sampling points in the direction of travel of the predetermined position; obstacle detection is performed based on the laser point cloud data of the region of interest; wherein the laser point cloud data is collected by a 3D laser radar; the region of interest is determined by performing the following operations: the path sampling points are sorted in order from near to far in the direction of travel of the predetermined position to form a path sampling point sequence; The predetermined position is taken as a starting point of the first region of interest At this time, ;​ Path sampling points are extracted sequentially from the path sampling point sequence. When the first sampling point is extracted... When sampling points along a path, perform a relative relationship determination: Acquisition North-bias deviation from the first path sampling point and straight-line distance , If , the first path sampling point is taken as the end point of the first region of interest , and the iteration ends. If and , update to , and extract the first path sampling point, and repeat the relative relationship judgment; If and , the first path sampling point is taken as the end point of the first region of interest ; the end point of the first region of interest is taken as the start point of the second region of interest , the second region of interest is established, and is updated to ; is updated to , the first path sampling point is extracted, and the relative relationship judgment is repeated; after the iteration is completed, the starting point and the ending point of one or more regions of interest are obtained, and the region of interest is determined based on the obtained starting point and ending point; wherein, represents a north-bias deviation threshold value, represents a cumulative distance threshold value, when , represents a straight-line distance between a start point and an end point of the th region of interest; , is a path sampling point sequence length. 2.The 3D lidar-based obstacle detection method for unmanned mine vehicles according to claim 1, wherein, The position of the path sampling point and the current position of the unmanned mine car each include a coordinate on an x-axis in a world coordinate system , a coordinate on a y-axis , and a northing angle . 3.The 3D lidar-based obstacle detection method for unmanned mine vehicles according to claim 2, characterized in that, the region of interest is determined based on the obtained starting point and ending point, comprising: the starting point and the ending point of each region of interest are dilated by a road width to obtain corner points on both sides of the starting point and the ending point, and the starting point and the ending point on the same side of each region of interest, the corner points on both sides of the starting point, the corner points on the other side of the starting point and the ending point, and the corner points on both sides of the ending point are sequentially connected to obtain a complete region of interest; a plurality of complete regions of interest are integrated to determine the region of interest. 4.The 3D lidar-based obstacle detection method for unmanned mine vehicles according to claim 3, wherein, The cumulative distance threshold is determined according to the predetermined distance, the driving speed of the unmanned mine vehicle and the detection distance of the 3D laser radar. 5.The 3D lidar-based obstacle detection method for unmanned mine vehicles according to claim 4, characterized in that, The cumulative distance threshold is 30 m, and the deviation of the north angle threshold is 5°. 6.The 3D lidar-based obstacle detection method for unmanned mine vehicles according to claim 5, wherein, A plurality of path sampling points are uniformly preset on the fixed driving track, and the distance between two adjacent path sampling points is 0.2 m.

7. The 3D lidar-based obstacle detection method for unmanned mine vehicles according to any one of claims 1-6, wherein, Obstacle detection is performed based on the laser point cloud data of the region of interest, comprising: the laser point cloud data of the region of interest is processed to obtain non-ground point cloud data in the region of interest; obstacle detection is performed on the non-ground point cloud data in the region of interest. 8.The 3D lidar-based obstacle detection method for unmanned mine vehicles according to claim 7, wherein, The non-ground point cloud data in the region of interest is obtained by: the laser point cloud data of the region of interest is filtered based on a voxel filtering method; the filtered laser point cloud data is segmented based on a line bundle ground segmentation method to obtain the non-ground point cloud data in the region of interest. 9.The 3D lidar-based obstacle detection method for unmanned mine vehicle according to claim 7 or 8, characterized in that, The non-ground point cloud data in the region of interest is detected by performing the following operations: Step S1: randomly select one non-ground point cloud data and label it as a class, and obtain the clustering radius corresponding to the non-ground point cloud data according to formula (1), （1） wherein is the horizontal angular resolution of the 3D lidar, is a variable threshold parameter, is the distance between the non-ground point cloud data and the center point of the 3D lidar; the non-ground point cloud data is taken as a clustering center, and the clustering radius corresponding to the non-ground point cloud data is taken as the clustering radius corresponding to the clustering center. Step S2: judging whether there is data meeting the clustering requirement in all unclassified non-ground point cloud data, if there is, dividing the non-ground point cloud data meeting the clustering requirement into the category to which the clustering center belongs, taking each non-ground point cloud data newly divided into the category as an updated clustering center respectively, and obtaining the clustering radius corresponding to the updated clustering center, repeating step S2 until there is no non-ground point cloud data meeting the clustering requirement; and segmenting all non-ground point cloud data meeting the clustering requirement into the target of the category; Step S3: judging whether there is unclassified non-ground point cloud data, if there is not, ending; if there is, entering step S4; Step S4: selecting an unclassified non-ground point cloud data, labeling a new category for the unclassified non-ground point cloud data, taking the unclassified non-ground point cloud data as an updated clustering center, obtaining the clustering radius corresponding to the updated clustering center, and then executing step S2; until all non-ground point cloud data are segmented into the target of a category; The clustering requirement refers to that the Euclidean distance between the unclassified non-ground point cloud data and the clustering center is less than the clustering radius corresponding to the clustering center; Step S5: judging the number of non-ground point cloud data contained in the target of each category, and detecting the target of the category meeting the number threshold requirement as an obstacle.

Images