A Fast Detection Method for Mining Area Obstacles Based on Multidimensional Trees

Abstract

The present invention relates to a method for quickly detecting obstacles in a mining area based on a multi-dimensional tree, belonging to the technical field of obstacle detection, and solving the problems of serious over-detection, high missed detection rate, and poor real-time performance in the process of obstacle fault detection in the mining area. The method includes: presetting a plurality of path sampling points on the fixed driving track of the mining operation; during the driving process of the driverless mining vehicle along the driving track, periodically collecting laser point cloud data and the current position of the driverless mining vehicle; every time a set of laser point cloud data and the current position of the driverless mining vehicle are collected periodically, obstacle detection is performed: determining the region of interest based on the current position of the driverless mining vehicle and the path sampling points; processing the laser point cloud data in the region of interest to obtain the corresponding non-ground point cloud data; constructing a multi-dimensional tree data structure of the non-ground point cloud data based on the covariance of the non-ground point cloud data in the three dimensions of x, y, and z; and performing obstacle detection based on the constructed multi-dimensional tree data structure.

Description

Technical Field

[0001] The present invention relates to the technical field of obstacle detection, and in particular, to a method for quickly detecting obstacles in a mining area based on a multi-dimensional tree. Background Art

[0002] The road environment in open-pit mines is complex, often accompanied by traffic factors such as soil slopes, deep pits, retaining walls, pedestrians, and other auxiliary operation vehicles. This poses high requirements for the safety of unmanned transportation operations of mining trucks. In recent years, with the construction of informatization and intelligentization in mining areas, more intelligent unmanned truck systems are being continuously applied to production operations. Among them, the environmental intelligent perception system, as an important part of the unmanned mining truck, mainly ensures the safe and reliable driving of the unmanned mining truck in the mining area environment, undertakes the task of effectively detecting pedestrians, vehicles, and other ground obstacles, and timely transmitting obstacle information to the decision-making system.

[0003] For obstacle detection on mining roads, extracting the region of interest is a relatively crucial step. It clips the original point cloud and only retains the point cloud data on the driving path, which can effectively improve the operation efficiency and reduce the missed detection rate. Traditional methods use the method of curb detection to clip the obtained point cloud. However, in mining areas, there are often flat roads without curbs, and this method will result in too much drivable road surface, causing over-detection and being unfavorable for improving the operation efficiency. There is also a method of point cloud segmentation based on the Z-axis coordinate of the laser point cloud, which requires detailed classification of the obstacle height and is prone to false detection. In terms of obstacle detection algorithms, there have been proposed the grid map method of compressing three-dimensional data into a two-dimensional plane and performing clustering, the feature image method of taking screenshots of the point cloud frame by frame and using visual methods for detection, etc. However, these methods can only obtain the target information on a certain two-dimensional plane, and in the actual application process, they often cannot effectively describe the object. Another type of detection method is to aggregate scattered data points into independent classifications based on the differences generated by a certain pre-determined criterion (such as density, distance, hierarchy, etc.). However, this type of clustering algorithm needs to calculate the distance between each adjacent point, which will generate a large time cost and does not meet the real-time requirement. Summary of the Invention

[0004] In view of the above analysis, the embodiments of the present invention aim to provide a method for quickly detecting obstacles in a mining area based on a multi-dimensional tree, so as to solve the problems of serious over-detection, high missed detection rate, and poor real-time performance existing in the existing obstacle fault detection process in mining areas.

[0005] The embodiments of the present invention provide a method for quickly detecting obstacles in a mining area based on a multi-dimensional tree, including:

[0006] Pre-set a plurality of path sampling points on the fixed driving track during mining operations; during the driving process of the driverless mining truck along the driving track, periodically collect the laser point cloud data and the current position of the driverless mining truck;

[0007] For each set of periodically collected laser point cloud data and the current position of the driverless mining truck, the following method is used for obstacle detection:

[0008] Based on the current position of the driverless mining truck and the path sampling points, determine the region of interest;

[0009] Process the laser point cloud data within the region of interest to obtain the corresponding non-ground point cloud data;

[0010] Based on the covariance of the non-ground point cloud data in the three dimensions of x, y, and z, construct a multi-dimensional tree data structure of the non-ground point cloud data;

[0011] Based on the constructed multi-dimensional tree data structure of the non-ground point cloud data, perform obstacle detection.

[0012] On the basis of the above solution, the present invention has also made the following improvements:

[0013] Furthermore, the construction of the multi-dimensional tree data structure of the non-ground point cloud data includes:

[0014] Calculate the covariance of the non-ground point cloud data in the three dimensions of x, y, and z respectively based on the coordinate values of all non-ground point cloud data in the three dimensions of x, y, and z;

[0015] Perform splitting in the dimension with the largest covariance, select the median point of all non-ground point cloud data in the dimension with the largest covariance, and mount the non-ground point cloud data corresponding to the median point to the root of the multi-dimensional tree; and use the non-ground point cloud data of all non-ground point cloud data in the dimension with the largest covariance that is less than the median point as the left subtree of the non-ground point cloud data at the root and the non-ground point cloud data greater than the median point as the right subtree of the non-ground point cloud data at the root;

[0016] For all non-ground point cloud data on the left and right subtrees, repeat the calculation of the dimension with the largest covariance and perform the mounting of the next-level non-ground point cloud data on the multi-dimensional tree according to the calculation results;

[0017] Repeat the above process until each non-ground point cloud data is mounted on the multi-dimensional tree, and construct the multi-dimensional tree data structure of the non-ground point cloud data.

[0018] Furthermore, perform obstacle detection by executing the following operations:

[0019] Step S1: Arbitrarily select a non-ground point cloud data in the multi-dimensional tree data structure, label it as a category, use this non-ground point cloud data as the clustering center, and use the clustering radius corresponding to this non-ground point cloud data as the clustering radius corresponding to this clustering center;

[0020] Step S2: Sequentially determine whether each unclassified non-ground point cloud data connected to this clustering center meets the clustering radius requirement. For the unclassified non-ground point cloud data that meets the clustering radius requirement, execute Step S3;

[0021] Step S3: Divide the unclassified non-ground point cloud data that meets the clustering radius requirement into the category to which this clustering center belongs, and obtain all unclassified non-ground point cloud data connected to the non-ground point cloud data that meets the clustering radius requirement; sequentially determine whether the unclassified non-ground point cloud data meets the clustering radius requirement. For the unclassified non-ground point cloud data that meets the clustering radius requirement, repeat Step S3 until there is no unclassified non-ground point cloud data that meets the clustering radius requirement;

[0022] Step S4: Use each non-ground point cloud data divided into this category as the updated clustering center respectively, and obtain the clustering radius corresponding to the updated clustering center, then transfer to Step S2 until there is no unclassified non-ground point cloud data that meets the clustering radius requirement;

[0023] Step S5: Determine whether there is unclassified non-ground point cloud data. If not, end; if so, enter Step S6;

[0024] Step S6: Arbitrarily select an unclassified non-ground point cloud data, label it with a new category, use this unclassified non-ground point cloud data as the updated clustering center, obtain the clustering radius corresponding to the updated clustering center, and then transfer to Step S2; until all non-ground point cloud data are segmented into the target of a certain category;

[0025] Step S7: Determine the number of non-ground point cloud data included in the targets of each category, and detect the targets of the categories that meet the number threshold requirement as obstacles.

[0026] Furthermore, obtain the clustering radius corresponding to each non-ground point cloud data according to formula (1),

[0027] (1)

[0028] In the formula, is the horizontal angular resolution of the 3D lidar, is a variable threshold parameter, is the distance between this non-ground point cloud data and the center point of the 3D lidar.

[0029] Further, the requirement for the clustering radius is that the Euclidean distance between the unclassified non-ground point cloud data and the clustering center does not exceed the clustering radius corresponding to the clustering center.

[0030] Further, determining the region of interest based on the current position of the unmanned mining vehicle and the path sampling points includes:

[0031] Taking the first path sampling point after the unmanned mining vehicle travels a predetermined distance along the driving trajectory from the current position as the predetermined position, and determining the region of interest based on the deviation angle northward and the distance between the predetermined position and the path sampling points in the traveling direction of the predetermined position.

[0032] Further, the region of interest is determined by performing the following operations:

[0033] Sort the path sampling points in the order from near to far in the traveling direction of the predetermined position to form a path sampling point sequence;

[0034] Taking the predetermined position as the starting point of the th region of interest , at this time, ;

[0035] Successively extract path sampling points from the path sampling point sequence. When the th path sampling point is extracted, perform a relative relationship judgment:

[0036] Obtain the deviation angle northward and the straight-line distance with the th path sampling point,

[0037] If , then take the th path sampling point as the end point of the th region of interest , and the iteration ends;

[0038] If and , update to , extract the th path sampling point, and repeat the relative relationship judgment;

[0039] If and , then take the th path sampling point as the end point of the th region of interest ; taking the end point of the th region of interest as the The starting point of a region of interest , establish the th region of interest, update to ; update to , and extract the th path sampling point, and repeat the relative relationship judgment;

[0040] After the iteration ends, obtain the starting point and ending point of one or more regions of interest, and based on the obtained starting point and ending point, determine the region of interest;

[0041] Among them, represents the deviation threshold of the northward angle, represents the cumulative distance threshold, , when , represents the straight-line distance between the starting point and the ending point of the th region of interest; , is the length of the path sampling point sequence.

[0042] Furthermore, based on the obtained starting point and ending point, determine the region of interest, including:

[0043] Dilate the starting point and ending point of each region of interest with the road width to obtain the corner points on both sides of the starting point and ending point, and sequentially connect the two corner points on the same side of the starting point and ending point 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 point and ending point, and the corner points on both sides of the ending point to obtain the complete region of interest;

[0044] Integrate multiple complete regions of interest and determine them as the region of interest.

[0045] Furthermore, determine the cumulative distance threshold according to the predetermined distance, the driving speed of the unmanned mining vehicle, and the detection distance of the 3D lidar.

[0046] Furthermore, process the laser point cloud data in the region of interest to obtain the corresponding non-ground point cloud data, including:

[0047] Filter the laser point cloud data in the region of interest based on the voxel filtering method;

[0048] Perform ground segmentation on the filtered laser point cloud data based on the ground segmentation method of the wire bundle to obtain the non-ground point cloud data in the region of interest.

[0049] Compared with the prior art, the present invention can at least achieve one of the following beneficial effects:

[0050] The method for quickly detecting obstacles in a mining area based on a multi-dimensional tree disclosed in the present invention can dynamically determine the region of interest according to the path sampling points on the fixed driving track of mining operations and the current position of the driverless mining vehicle, and perform obstacle detection on the laser point cloud data within the dynamically determined region of interest. This method effectively solves the problems of missing road edges caused by irregular road surfaces in the mining area and over-detection of obstacles due to overly wide road surfaces;

[0051] Meanwhile, before obstacle detection, by calculating the covariance of the non-ground point cloud data within the region of interest in each dimension, a multi-dimensional tree data structure of the non-ground point cloud data is constructed based on the dimension with the largest covariance. The multi-dimensional tree data structure constructed in this way can well represent the discrete situation among the non-ground point cloud data;

[0052] In addition, during the obstacle detection process, the clustering center is first located in the constructed multi-dimensional tree data structure of the non-ground point cloud data, and by determining whether the unclassified non-ground point cloud data directly associated with the clustering center meets the clustering radius requirement, only when the clustering radius requirement is met, the judgment is extended to the clustering radius requirement judgment of the adjacent unclassified non-ground point cloud data, otherwise the judgment process is stopped. Since the multi-dimensional tree data structure can well represent the discrete situation among the non-ground point cloud data, if the non-ground point cloud data adjacent to the clustering center do not meet the clustering radius requirement, it can be determined that other non-adjacent non-ground point cloud data are even less likely to meet the clustering radius requirement; this method can effectively reduce the data to be traversed during the clustering process and improve the speed of obstacle detection at the same time;

[0053] Finally, during the obstacle detection process, this embodiment can also dynamically update the clustering radius according to the distance between the non-ground point cloud data and the center point of the 3D lidar, so as to well meet the detection process of obstacles at different distances from the driverless mining vehicle, effectively reducing the missed detection rate.

[0054] In the present invention, the above technical solutions can also be combined with each other to achieve more preferred combination schemes. Other features and advantages of the present invention will be described in the subsequent specification, and some advantages can be made obvious from the specification, or understood by implementing the present invention. The objectives and other advantages of the present invention can be realized and obtained through the content specifically pointed out in the specification and the drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] The drawings are only for the purpose of showing specific embodiments and are not considered to be a limitation of the present invention. Throughout the drawings, the same reference signs represent the same components;

[0056] Figure 1 It is a flowchart of the method for quickly detecting obstacles in a mining area based on a multi-dimensional tree in an embodiment of the present invention;

[0057] Figure 2 This is an example of a fixed driving trajectory in an embodiment of the present invention;

[0058] Figure 3 (a) Figure 3 (b) are respectively laser point cloud data distribution diagrams before and after the region of interest corresponding to the straight-moving vehicle trajectory is determined in an embodiment of the present invention;

[0059] Figure 4 (a) Figure 4 (b) are respectively the laser point cloud data distribution diagrams before and after the region of interest corresponding to the curved driving trajectory in the embodiment of the present invention is determined;

[0060] Figure 5 This is an example of an original mining scene in an embodiment of the present invention;

[0061] Figure 6 (a) Figure 6 (b) are point cloud effect diagrams before and after ground segmentation in the embodiments of the present invention;

[0062] Figure 7 This is an example of a multidimensional tree data structure in an embodiment of the present invention. DETAILED DESCRIPTION

[0063] The preferred embodiments of the present invention are described in detail below in conjunction with the accompanying drawings, wherein the accompanying drawings constitute a 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 used to limit the scope of the present invention.

[0064] A specific embodiment of the present invention discloses a method for rapid detection of obstacles in a mining area based on a multidimensional tree, the flow chart of which is as follows: Figure 1 As shown, the following steps are included:

[0065] Step S1: Preset multiple path sampling points on a fixed driving track of the mining area; periodically collect laser point cloud data and the current position of the unmanned mining vehicle while the unmanned mining vehicle is driving along the driving track;

[0066] In step S1, the 3D lidar is used to collect the lidar point cloud data. The horizontal field of view of the lidar is 360°, and the vertical field of view is 45° (±22.5°). During the use process, a lot of redundant point cloud data will be generated, and these redundant point cloud data will greatly reduce the calculation speed and affect the real-time performance of the entire method. Since during the operation in the mining area, only the obstacles that appear within a certain range centered on the driving trajectory will affect the driving process of the unmanned mining vehicle. Therefore, it is only necessary to determine whether there are obstacles within this range based on the point cloud data within this range. This embodiment proposes a solution for dynamically extracting the region of interest from the collected point cloud and dynamically retaining the point cloud data within the drivable region; the specific process is as shown in step S2:

[0067] Step S2: Every time a set of lidar point cloud data and the current position of the unmanned mining vehicle are periodically collected, the following method is used for obstacle detection:

[0068] Step S21: Based on the current position of the unmanned mining vehicle and the path sampling points, determine the region of interest; specifically,

[0069] Take the first path sampling point after the unmanned mining vehicle travels a predetermined distance along the driving trajectory from the current position as the predetermined position, and determine the region of interest based on the northerly angle deviation and distance between the predetermined position and the path sampling points in the traveling direction of the predetermined position.

[0070] It should be noted that the obstacle detection of the unmanned mining vehicle must be predictive, that is, it can determine the obstacles on a future driving path in advance, so that the unmanned mining vehicle can make a reaction in advance according to the obstacle detection result. Therefore, this embodiment selects the first path sampling point after the unmanned mining vehicle travels a predetermined distance along the driving trajectory from the current position as the predetermined position, and determines the region of interest based on the northerly angle deviation and distance between the predetermined position and the path sampling points in the traveling direction of the predetermined position. In the actual application process, the predetermined distance is preferably 4m.

[0071] Preferably, in this embodiment, the positions of the path sampling points and the current position of the unmanned mining vehicle both include: the coordinates on the x-axis in the world coordinate system and the coordinates on the y-axis , as well as the northerly angle .

[0072] To ensure the determination accuracy of the region of interest, in this embodiment, a plurality of path sampling points are uniformly preset on the fixed driving trajectory, and the distance between adjacent two path sampling points is 0.2m.

[0073] In step S21, the region of interest is determined by the following method:

[0074] Sort the path sampling points in the order from near to far with respect to the traveling direction of the predetermined position to form a sequence of path sampling points;

[0075] Take the predetermined position as the starting point of the th region of interest , at this time, ;

[0076] Extract path sampling points from the sequence of path sampling points in turn. When the th path sampling point is extracted, perform a relative relationship judgment:

[0077] Obtain the deviation of the northward angle and the straight-line distance from the th path sampling point ,

[0078] If , then take the th path sampling point as the ending point of the th region of interest , and the iteration ends;

[0079] If and , update to , extract the th path sampling point, and repeat the relative relationship judgment;

[0080] If and , then take the th path sampling point as the ending point of the th region of interest ; Take the ending point of the th region of interest as the starting point of the th region of interest , establish the th region of interest, update to ; Update to , extract the th path sampling point, and repeat the relative relationship judgment;

[0081] After the iteration ends, obtain the starting points and ending points of one or more regions of interest, and determine the regions of interest based on the obtained starting points and ending points;

[0082] Among them, represents the northward angle deviation threshold, represents the cumulative distance threshold, , when , represents the straight-line distance between the starting point and the ending point of the th region of interest; , is the length of the path sampling point sequence.

[0083] Preferably, the starting point and the ending point of each region of interest are dilated with the road width to obtain the corner points on both sides of the starting point and the ending point. The two corner points on the same side of the starting point and the ending point 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 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; multiple complete regions of interest are integrated and determined as the region of interest. In the actual implementation of this solution, the road width is preferably 4m.

[0084] Preferably, in this embodiment, the cumulative distance threshold is determined according to the predetermined distance, the driving speed of the driverless mining truck, and the detection distance of the 3D lidar. Preferably, the cumulative distance threshold is 30m; the deviation threshold of the northward angle is 5°.

[0085] Figure 3 in (a), Figure 3 in (b) are the laser point cloud data distribution diagrams before and after determining the region of interest corresponding to the straight driving trajectory; Figure 4 in (a), Figure 4 in (b) are the laser point cloud data distribution diagrams before and after determining the region of interest corresponding to the curved driving trajectory; It can be analyzed that by determining the region of interest, the data volume of the laser point cloud data can be greatly reduced, thereby effectively reducing the laser point cloud data required for subsequent work and improving the work efficiency.

[0086] Step S22: Process the laser point cloud data in the region of interest to obtain the corresponding non-ground point cloud data;

[0087] Filter the laser point cloud data in the region of interest based on the voxel filtering method;

[0088] Perform ground segmentation on the filtered laser point cloud data based on the ground segmentation method of the wire bundle to obtain the non-ground point cloud data in the region of interest.

[0089] Specifically, voxel filtering is a commonly used filtering method for three-dimensional point cloud data, which can reduce the number of point clouds for operation while maintaining the characteristics of each part of the point cloud. Assume that the point cloud space is in a cuboid, and the point cloud space is cut with cuboid units of a fixed size. The dimensions of the cuboid unit along the axis are respectively .

[0090] The point cloud data within the region of interest contains not only the points of obstacles but also a large number of ground points. However, the numerous ground points will interfere with obstacle detection. Therefore, it is necessary to remove the ground points. In this embodiment, a ground segmentation method based on the beam 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, and the distance between the point cloud data below the horizontal line of the lidar and the lidar in the horizontal direction at the same angle is calculated according to formula (1), and the point cloud data is sorted in ascending order of the distance;

[0091] (1)

[0092] 、 respectively represent the coordinate values of the point cloud data on the 、 axes;

[0093] After obtaining the sorted point cloud data at each angle, for each point cloud data, it is judged whether the slope between the point cloud data and the point cloud closest to it is less than or equal to the slope threshold. If so, the point cloud data is judged as a ground point;

[0094] Otherwise, it is further judged whether the slope between the point cloud data and the center point of the lidar exceeds the road surface threshold. If it exceeds, the point cloud data is judged as a ground point; otherwise, it is judged as a non-ground point.

[0095] The slope threshold and the road surface threshold can be adaptively set based on the overall flatness of the road surface in the mining area;

[0096] Among them, the formula for calculating the slope between two point cloud data is:

[0097] (2)

[0098] In the formula, represents the height difference between adjacent two points, represents the distance difference between adjacent two points. The original scene is as shown in Figure 5 . The point cloud effect diagrams before and after ground segmentation are respectively as shown in (a) in Figure 6 and (b) in Figure 6 . Step S23: Construct a multi-dimensional tree data structure of the non-ground point cloud data based on the covariance of the non-ground point cloud data in the x, y, and z dimensions;

[0099] In this embodiment, a method of constructing a multi-dimensional tree is used to construct a data structure for subsequent clustering, which can effectively reduce the number of searches and improve the calculation efficiency. A multi-dimensional tree is a binary tree in which each node is a numerical point in a certain dimension. Each node on it represents a hyperplane that is perpendicular to the coordinate axis of the current division dimension and divides the space into two parts in this dimension, with one part in its left subtree and the other part in its right subtree. Moreover, the coordinate values of all points on the left subtree in this dimension are less than the coordinate value of the root node in the current dimension, and the coordinate values of all points on the right subtree in this dimension are greater than or equal to the coordinate value of the root node in the current dimension.

[0100] Calculate the covariance of the non-ground point cloud data in the x, y, and z dimensions respectively based on the coordinate values of all non-ground point cloud data in the x, y, and z dimensions;

[0101] Perform segmentation in the dimension with the largest covariance, select the median point of all non-ground point cloud data in the dimension with the largest covariance, and mount the non-ground point cloud data corresponding to this median point to the root of the multi-dimensional tree; and use the non-ground point cloud data that is less than the median point of all non-ground point cloud data in the dimension with the largest covariance as the left subtree of the non-ground point cloud data at the root, and the non-ground point cloud data that is greater than the median point as the right subtree of the non-ground point cloud data at the root; thus, two relatively balanced arrays can be obtained, and the point cloud data distributions in these two arrays are basically the same.

[0102] For all non-ground point cloud data on the left and right subtrees, repeat the calculation of the dimension with the largest covariance, and perform the mounting of the next-level non-ground point cloud data on the multi-dimensional tree according to the calculation results;

[0103] Repeat the above process until each non-ground point cloud data is mounted on the multi-dimensional tree, and construct the multi-dimensional tree data structure of the non-ground point cloud data.

[0104] Figure 7 Is an example of a multi-dimensional tree data structure. In Figure 7 The multi-dimensional tree data dimension in the example is 3D, and the multi-dimensional tree is mounted in the order of x->y->z->x->y, and finally all points are mounted on the multi-dimensional tree.

[0105] Step S24: Perform obstacle detection based on the constructed multi-dimensional tree data structure of the non-ground point cloud data.

[0106] Obstacle detection is the core content of driverless mining. The key issues of obstacle detection lie in the selection of the clustering radius and the determination of obstacle criteria. To more effectively identify obstacles, due to the different distribution positions of targets, the number of laser point cloud data distributed on different obstacles is also different: for obstacles at a relatively far distance, the number of distributed point clouds is relatively small; for obstacles at a relatively close distance, the number of distributed point clouds is relatively large. Therefore, this paper uses a method based on a variable clustering radius to detect obstacles in the region of interest. Specifically:

[0107] Step S241: Arbitrarily select a non-ground point cloud data in the multi-dimensional tree data structure, label it as a category, use this non-ground point cloud data as the clustering center, and use the clustering radius corresponding to this non-ground point cloud data as the clustering radius corresponding to this clustering center;

[0108] Preferably, obtain the clustering radius corresponding to each non-ground point cloud data according to formula (3),

[0109] (3)

[0110] In the formula, is the horizontal angular resolution of the 3D lidar, is the variable threshold parameter, is the distance between this non-ground point cloud data and the center point of the 3D lidar.

[0111] Step S242: Sequentially determine whether each unclassified non-ground point cloud data connected to this clustering center meets the clustering radius requirement. For the unclassified non-ground point cloud data that meets the clustering radius requirement, execute step S243;

[0112] Preferably, the clustering radius requirement is that the Euclidean distance between the unclassified non-ground point cloud data and the clustering center does not exceed the clustering radius corresponding to this clustering center.

[0113] Step S243: Divide the unclassified non-ground point cloud data that meets the clustering radius requirement into the category to which this clustering center belongs, and obtain all unclassified non-ground point cloud data connected to the non-ground point cloud data that meets the clustering radius requirement; sequentially determine whether the unclassified non-ground point cloud data meets the clustering radius requirement. For the unclassified non-ground point cloud data that meets the clustering radius requirement, repeat step S243 until there is no unclassified non-ground point cloud data that meets the clustering radius requirement;

[0114] Step S244: Use each non-ground point cloud data divided into this category as the updated clustering center respectively, and obtain the clustering radius corresponding to the updated clustering center, then transfer to step S242 until there is no unclassified non-ground point cloud data that meets the clustering radius requirement;

[0115] Step S245: Determine whether there is unclassified non-ground point cloud data. If not, end the process; if so, proceed to Step S246;

[0116] Step S246: Arbitrarily select an unclassified non-ground point cloud data, label it with a new category, and use this unclassified non-ground point cloud data as the updated clustering center. Obtain the clustering radius corresponding to the updated clustering center, and then transfer to Step S242; until all non-ground point cloud data are segmented into targets of a certain category;

[0117] Step S247: Determine the number of non-ground point cloud data included in the targets of each category, and detect the targets of the category that meet the requirement of the number threshold as obstacles. If the number of non-ground point cloud data included in a certain category of target is small, it indicates that the target of this category may be noise or miscellaneous points; while when the number of non-ground point cloud data included in a certain category of target is large, it can indicate that the target of this category is an obstacle. By performing the above process, one or more obstacles can be determined. After repeated experiments, the preferred number threshold is 4. At this time, it is possible to well judge whether the targets of each category are obstacles, thus completing the obstacle detection process.

[0118] During the obstacle detection process, by using the multi-dimensional tree data structure, the search speed of the three-dimensional point cloud data is accelerated, the laser point cloud information is fully utilized, obstacles such as mine dumps and retaining walls in the mining area are detected, and the detection efficiency is ensured. At the same time, for each clustering center, the clustering radius corresponding to the clustering center is dynamically adjusted according to formula (4), which can effectively improve the problem of inconsistent clustering radii caused by the distance of obstacles, thereby more effectively identifying obstacles.

[0119] In summary, the fast mining area obstacle detection method based on the multi-dimensional tree provided in this embodiment can dynamically determine the region of interest according to the path sampling points on the fixed driving trajectory of the mining operation and the current position of the unmanned mining vehicle, and perform obstacle detection on the laser point cloud data in the dynamically determined region of interest. This method effectively solves the problems of missing road edges caused by irregular road surfaces in the mining area and over-detection of obstacles due to overly wide road surfaces;

[0120] At the same time, before obstacle detection, by calculating the covariance of the non-ground point cloud data in the region of interest in each dimension, a multi-dimensional tree data structure of the non-ground point cloud data is constructed based on the dimension with the largest covariance. The multi-dimensional tree data structure constructed in this way can well represent the discrete situation between the non-ground point cloud data;

[0121] In addition, during the obstacle detection process, in the multi-dimensional tree data structure of the constructed non-ground point cloud data, the clustering center is located first, and by determining whether the unclassified non-ground point cloud data directly associated with the clustering center meets the clustering radius requirement, the judgment process of the clustering radius requirement of the adjacent unclassified non-ground point cloud data is only spread when the clustering radius requirement is met, otherwise the judgment process is stopped. Since the multi-dimensional tree data structure can well represent the discrete situation between non-ground point cloud data, if the non-ground point cloud data adjacent to the clustering center do not meet the clustering radius requirement, it is even more impossible for other non-adjacent non-ground point cloud data to meet the clustering radius requirement; this method can effectively reduce the data to be traversed during the clustering process and improve the speed of obstacle detection at the same time;

[0122] Finally, during the obstacle detection process, this embodiment can also dynamically update the clustering radius according to the distance between the non-ground point cloud data and the center point of the 3D lidar, so as to well meet the detection process of obstacles at different distances from the unmanned mining vehicle, effectively reducing the missed detection rate.

[0123] Those skilled in the art can understand that all or part of the processes of implementing the above method embodiments can be completed by instructing relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium. Among them, the computer-readable storage medium is a magnetic disk, an optical disk, a read-only memory or a random access memory, etc.

[0124] The above is only a specific and preferred embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed by the present invention should be covered by the protection scope of the present invention.

Claims

1. A fast detection method for mining area obstacles based on a multi-dimensional tree, characterized in that Including: Pre-setting a plurality of path sampling points on a fixed driving track during mining operations in a mining area; During the driving process of an unmanned mining truck along the driving track, periodically collecting laser point cloud data and the current position of the unmanned mining truck; For each periodic collection of a set of laser point cloud data and the current position of the unmanned mining truck, the following method is used for obstacle detection: Based on the current position of the unmanned mining truck and the path sampling points, determining an area of interest; Processing the laser point cloud data within the area of interest to obtain corresponding non-ground point cloud data; Based on the covariance of the non-ground point cloud data in the three dimensions of x, y, and z, constructing a multi-dimensional tree data structure of the non-ground point cloud data; Based on the constructed multi-dimensional tree data structure of the non-ground point cloud data, performing obstacle detection; The constructing of the multi-dimensional tree data structure of the non-ground point cloud data includes: Respectively calculating the covariance of the non-ground point cloud data in the three dimensions of x, y, and z based on the coordinate values of all the non-ground point cloud data in the three dimensions of x, y, and z; Taking the dimension with the largest covariance for splitting, selecting the median point of all the non-ground point cloud data in the dimension with the largest covariance, and mounting the non-ground point cloud data corresponding to the median point to the root of the multi-dimensional tree; and taking the non-ground point cloud data of all the non-ground point cloud data in the dimension with the largest covariance that is less than the median point as the left subtree of the non-ground point cloud data at the root and the non-ground point cloud data that is greater than the median point as the right subtree of the non-ground point cloud data at the root; For all the non-ground point cloud data on the left and right subtrees, repeating the calculation of the dimension with the largest covariance and performing the mounting of the next-level non-ground point cloud data on the multi-dimensional tree according to the calculation results; Repeating the above process until each non-ground point cloud data is mounted on the multi-dimensional tree, constructing the multi-dimensional tree data structure of the non-ground point cloud data.

2. The method for quickly detecting obstacles in a mining area based on a multi-dimensional tree according to claim 1, characterized in that, Performing obstacle detection by executing the following operations: Step S1: Arbitrarily taking a non-ground point cloud data in the multi-dimensional tree data structure, labeling it as a category, taking this non-ground point cloud data as a clustering center, and taking the clustering radius corresponding to this non-ground point cloud data as the clustering radius corresponding to this clustering center; Step S2: Sequentially determining whether each unclassified non-ground point cloud data connected to this clustering center meets the clustering radius requirement. For the unclassified non-ground point cloud data that meets the clustering radius requirement, execute Step S3; Step S3: Dividing the unclassified non-ground point cloud data that meets the clustering radius requirement into the category to which this clustering center belongs, and obtaining all the unclassified non-ground point cloud data connected to the non-ground point cloud data that meets the clustering radius requirement; sequentially determining whether the unclassified non-ground point cloud data meets the clustering radius requirement. For the unclassified non-ground point cloud data that meets the clustering radius requirement, repeating Step S3 until there is no unclassified non-ground point cloud data that meets the clustering radius requirement; Step S4: Taking each non-ground point cloud data divided into this category as an updated clustering center respectively and obtaining the clustering radius corresponding to the updated clustering center, and transferring to Step S2 until there is no unclassified non-ground point cloud data that meets the clustering radius requirement; Step S5: Determine whether there is unclassified non-ground point cloud data. If not, end; if so, proceed to Step S6; Step S6: Arbitrarily select an unclassified non-ground point cloud data, label it with a new category, and use this unclassified non-ground point cloud data as the updated clustering center. Obtain the clustering radius corresponding to the updated clustering center, and then transfer to Step S2; until all non-ground point cloud data are segmented into the targets of a certain category; Step S7: Determine the number of non-ground point cloud data included in the targets of each category, and detect the targets of the categories that meet the number threshold requirement as obstacles.

3. The method for quickly detecting obstacles in a mining area based on a multi-dimensional tree according to claim 2, wherein Obtain the clustering radius corresponding to each non-ground point cloud data according to formula (1), （1） In the formula, is the horizontal angular resolution of the 3D lidar, is the variable threshold parameter, is the distance between the non-ground point cloud data and the center point of the 3D lidar.

4. The method for rapid detection of obstacles in a mining area based on a multi-dimensional tree according to claim 3, wherein, The requirement for the clustering radius is that the Euclidean distance between the unclassified non-ground point cloud data and the clustering center does not exceed the clustering radius corresponding to this clustering center.

5. The method for rapid detection of obstacles in a mining area based on a multi-dimensional tree according to any one of claims 1-4, characterized in that Determine the region of interest based on the current position of the unmanned mining vehicle and the path sampling points, including: Take the first path sampling point after the unmanned mining vehicle travels a predetermined distance along the driving trajectory from the current position as the predetermined position, and determine the region of interest based on the deviation angle and distance in the north direction between the predetermined position and the path sampling points in the traveling direction of the predetermined position.

6. The method for rapid detection of obstacles in a mining area based on a multi-dimensional tree according to claim 5, characterized in that Determine the region of interest by performing the following operations: Sort the path sampling points in the order from near to far in the traveling direction of the predetermined position to form a sequence of path sampling points; Taking a predetermined position as the starting point of the th region of interest, at this time, ; Extract path sampling points from the path sampling point sequence in sequence. When the th path sampling point is extracted, perform relative relationship judgment: Obtain with the northward angle deviation of the th path sampling point and the linear distance If , then use the th path sampling point as the end point of the th region of interest , and end the iteration; If and , update to , and extract the th path sampling point, and repeat the relative relationship judgment; If and , then use the th path sampling point as the termination point of the th region of interest ; Use the termination point of the th region of interest as the starting point of the th region of interest , establish the th region of interest, and update to ; Update to , and extract the th path sampling point, and repeat the relative relationship judgment; After the iteration ends, obtain the starting points and ending points of one or more regions of interest, and determine the region of interest based on the obtained starting points and ending points; Among them, represents the north deviation angle threshold, represents the cumulative distance threshold, , when , represents the straight-line distance between the starting point and the ending point of the th region of interest; , is the length of the path sampling point sequence.

7. The method for rapid detection of obstacles in a mining area based on a multi-dimensional tree according to claim 6, characterized in that, Determine the region of interest based on the obtained starting points and ending points, including: Dilate the starting points and ending points of each region of interest with the road width to obtain the corner points on both sides of the starting points and ending points. Connect the two corner points on the same side of the starting point and ending point 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 point and ending point, and the corner points on both sides of the ending point in sequence to obtain the complete region of interest; Integrate multiple complete regions of interest and determine them as the region of interest.

8. The method for rapid detection of obstacles in a mining area based on a multi-dimensional tree according to claim 7, characterized in that, Determine the cumulative distance threshold according to the predetermined distance, the driving speed of the unmanned mining vehicle, and the detection distance of the 3D lidar.

9. The method for rapid detection of obstacles in a mining area based on a multi-dimensional tree according to claim 5, characterized in that Process the laser point cloud data in the region of interest to obtain the corresponding non-ground point cloud data, including: Perform filtering processing on the laser point cloud data in the region of interest based on the voxel filtering method; Perform ground segmentation on the filtered laser point cloud data based on the ground segmentation method of the wire bundle to obtain the non-ground point cloud data in the region of interest.

Images