CN111537994A - An unmanned mine card obstacle detection method - Google Patents

Abstract

The invention discloses an obstacle detection method for unmanned mining trucks, which works according to the following steps: transforming obstacle data obtained by laser radar and millimeter wave radar into corresponding vehicle body coordinate systems; adopting the grid map height difference combined field Differential ground detection, draw a 0, 1 binary map of "ground-elevated points"; use a multi-parameter model to cluster the elevated points; according to the motion trajectory of the vehicle, determine whether the clustering result will be an obstacle that affects the normal driving of the vehicle Detect whether the vehicle is in a drivable area; match the obstacle data obtained by the millimeter wave radar with the obstacle data obtained by the lidar, and output the final result. Aiming at the actual application environment of mining dump trucks, the invention can effectively detect obstacles in the road, prevent missed detection, and accurately cluster. The present invention has good robustness, adopts a variety of radar fusion schemes, and reduces the false detection rate by matching the detection result of the laser radar with the detection result of the millimeter wave radar.

Description

An unmanned mine card obstacle detection method

technical field

The invention relates to the technical field of unmanned mining trucks, in particular to an obstacle detection method for unmanned mining trucks.

Background technique

The mine scene has a single operation and a relatively simple scene, which is the best scene for the implementation of unmanned driving technology. Compared with urban roads, the main characteristics of obstacle detection in mine scenes are: the road is rough and bumpy, and the mining truck is always in a bumpy state during driving, which will affect the sensor data collection; road obstacles, in addition to vehicles and pedestrians, there will also be rockfalls The perception system needs to accurately detect obstacles of a certain size; the road has no obvious curb, and the sides of the road are mostly trees, weeds, and soil slopes, which increases the difficulty of detecting the drivable area. In addition, in the mine scene, the design of the perception system needs to consider harsh conditions such as rain and snow, strong and weak light, and dust, which puts forward higher requirements for obstacle detection.

There is a method for monitoring obstacles related to unmanned mine trucks in the prior art. The patent application number 201610687204.6 discloses a rapid detection method for obstacles. For general vehicles, the method of using point cloud grid map and internal height difference is used to detect obstacles. , it is easy to miss detection; the template matching method is used to cluster various obstacles on the road, but in the 3D scene, due to the constant change of the relative position of the lidar and the target, the clustering result and template matching calculation will be inaccurate The problem. Application No. 201110150818.8 discloses a method for detecting unstructured road boundaries by combining SVM and lidar. This patent is not perfect for the detection of structured roads. Due to the lack of obvious road edge information, its safety cannot be guaranteed.

SUMMARY OF THE INVENTION

In order to solve the deficiencies in the prior art, the present invention provides an unmanned mine card obstacle detection method, which solves the technical problems of the prior art that the unmanned mine card obstacle detection capability is not strong and the robustness is poor. .

In order to achieve the above goals, the present invention adopts the following technical solutions:

An unmanned mine card obstacle detection method is characterized in that: work according to the following steps:

Convert the obstacle data obtained by lidar and millimeter-wave radar into the corresponding vehicle body coordinate system respectively;

Using the grid map height difference combined with the field difference ground detection, draw the 0, 1 binary map of "ground-elevated point";

Use a multi-parameter model to cluster elevated points;

According to the motion trajectory of the vehicle, determine whether the clustering result will be an obstacle that affects the normal driving of the vehicle;

Detect whether the vehicle is in a drivable area;

Match the obstacle data obtained by the millimeter wave radar with the obstacle data obtained by the lidar, and output the final result.

As a preferred solution of the present invention, the aforementioned method for detecting obstacles in an unmanned mine card: using the height difference of the grid map combined with the ground detection of the field difference, the 0, 1 binary map of "ground-elevated point" is drawn. The method is: gridize the obstacle data, obtain an N*N grid map, calculate the difference between the maximum and minimum elevations of the point cloud in each grid, and apply a gradient to the difference of each grid Calculate and compare the difference with the preset threshold, identify the point with the difference greater than the threshold as "elevated point", and perform neighborhood difference calculation on the point cloud whose difference is less than the threshold, and separate out the "elevated point" Grid and Ground Grid to fit all ground rasters to the road surface.

As a preferred solution of the present invention, the aforementioned method for detecting obstacles in unmanned mine cards: the selection of the field in the calculation of the neighborhood difference is to identify the grid within 2 meters around the grid as the neighborhood.

As a preferred solution of the present invention, the aforementioned method for detecting obstacles in an unmanned mine truck: the method for separating the ground grid is: according to the distribution of vehicle distances, all point clouds belonging to the ground are divided into multiple point cloud intervals, The plane fitting is performed on each point cloud interval, and all the fitted planes are connected, that is, the road surface.

As a preferred solution of the present invention, the aforesaid method for detecting obstacles in unmanned mine cards: the multi-parameter model performs clustering on the elevated points using the octagonal domain method for clustering, and obtains the clusters of elevated points. Combine multiple elevated point clusters into one to complete the target extraction of point clouds.

As a preferred solution of the present invention, in the aforementioned method for detecting obstacles in unmanned mine cards, the merging parameters include the size and density of clusters of elevated points.

As a preferred solution of the present invention, the aforementioned method for detecting obstacles in unmanned mine cards: convert the clustered elevated point clusters into two-dimensional data, and calculate the convex hull information of the two-dimensional data. The package information and the driving trajectory of the vehicle are used to determine whether the safe driving of the vehicle is affected.

As a preferred solution of the present invention, the aforementioned method for detecting obstacles in unmanned mine cards: it is necessary to expand the left and right sides of the vehicle's running track by a certain width as the safety zone of the vehicle. If there are clusters of elevated points in the safety zone , it is considered that there is an obstacle in the safe area.

As a preferred solution of the present invention, the aforementioned method for detecting obstacles in an unmanned mine card: the method for detecting whether a vehicle is in a drivable area is: selecting a certain point in front of the vehicle as a seed point, and adopting the four-neighborhood growth method, Retain the non-obstacle area in front of the mine card, calculate the distance value between the center point in front of the vehicle and the non-obstacle area, sort the distance value data to obtain the minimum value, and compare the minimum value with the set threshold value, If the minimum value is less than the set threshold, the vehicle is adjusted, otherwise it is deemed that the vehicle is safe.

As a preferred solution of the present invention, the aforementioned method for detecting obstacles in an unmanned mine card: select the center point in front of the vehicle as the seed point.

Beneficial effects achieved by the present invention:

Compared with the prior art, aiming at the actual application environment of mining dump trucks, the present invention proposes an obstacle detection system for self-driving mining dump trucks, which can effectively detect obstacles in the road, prevent missed detection, and accurately. Clustering; effective detection of road edges to ensure safety, especially suitable for unstructured road environments such as mines. And it is also applicable under harsh conditions such as rain and snow, strong and weak light, and dust, and has good robustness. From the perspective of applicable scenarios, the detection target includes structured roads in addition to the mine scene (unstructured road), to prevent the mining card from driving into the space other than the road when the positioning information is invalid.

The present invention has good robustness. According to the working principle and advantages and disadvantages of the laser radar, the present invention also adopts a variety of radar fusion schemes. false detection rate. If the matching is successful, the lidar detection result is saved as the final detection result, and the distance, size, orientation and other information of the obstacles are retained. If the matching fails, the matching failures in the lidar obstacles are filtered out, and the matching is successful, and then the matching Successful output of valid obstacle information.

Description of drawings

Fig. 1 is the overall work flow chart of the present invention;

Fig. 2 is the flow chart of system ground detection of the present invention;

Fig. 3 is a flow chart of road information acquisition of the present invention.

Detailed ways

The present invention will be further described below in conjunction with the accompanying drawings. The following examples are only used to illustrate the technical solutions of the present invention more clearly, and cannot be used to limit the protection scope of the present invention.

This embodiment discloses a method for detecting obstacles in an unmanned mine card. Referring to FIG. 1 , the specific steps are as follows:

The obstacle data obtained by lidar and millimeter-wave radar are converted into corresponding vehicle body coordinate systems respectively. Since there are many types of lidars, and there may be more than one of each, data fusion of multiple sensors is required.

The detection of ground data uses multiple planes to fit the road surface, and uses the height difference of the grid map combined with the detection of the neighborhood difference. The traditional lidar processing solution requires the separation of ground points and elevated points. On structured roads (mainly including expressways and urban arterial roads), the road surface can be approximated as a plane, and the road surface can be extracted by the plane extraction method. On unstructured roads, where the road is rough, multiple planes are used to fit the road surface. In combination with the grid map height difference and the neighborhood difference detection method, the method of extracting through the plane in the prior art will have a large error.

However, the present embodiment adopts the grid map height difference combined with the neighborhood difference detection method, which can stably and effectively detect the ground.

Step 1. Use the height difference of the grid map combined with the ground detection of the neighborhood difference to draw a 0, 1 binary map of "ground-elevated point". First, the lidar point cloud data is rasterized to obtain an N*N raster image; then the difference between the maximum and minimum point cloud elevations in each raster is calculated, and its gradient is calculated, and the difference A is calculated. Compare with the preset threshold D1, filter out the points whose difference is greater than the threshold (A is greater than D1) as "elevated points", keep the points whose difference is less than the threshold (A is less than D1), and identify as the ground, for the difference The points smaller than the threshold (A is smaller than D1) are used for neighborhood difference calculation. The selection of the neighborhood is related to the size of the grid. The selection principle in this embodiment is to consider the grid within 2 meters around the grid as a "neighborhood"; The point cloud belonging to the ground is divided into multiple point cloud intervals according to the vehicle distance distribution, and plane fitting is performed on each point cloud interval, and the fitted planes are connected, which is the road surface. From this, a 0, 1 binary map of "ground-elevated point" is drawn.

In the second step, a multi-parameter model is used to cluster the elevated points. For the 0, 1 binary map of "ground-elevated point", the eight-connected domain method is used to cluster, and the cluster of elevated points is obtained. Due to the large size of the mine card in the mine scene and the fact that the outer contour is not a standard rectangle, the point cloud will be relatively sparse during lidar detection, and the target feature points will be occluded by itself. After clustering, a single target will be clustered into multiple clusters. Case. Using the post-processing method of the multi-parameter model, the multiple clusters are merged into one to complete the target extraction of the point cloud. According to the size and density of clusters, different merging parameters are set. According to this parameter, clusters of different sizes are merged into one, thereby solving the problem of inaccurate clustering caused by sparse point clouds or irregular targets.

The third step is to detect and determine obstacles in the safety detection area based on the vehicle's target driving route: convert the clustered clusters into two-dimensional data, calculate their convex hull (obstacle) information, and determine whether it is an obstacle according to the convex hull information. obstacle;

The driving track of the vehicle is imported into the system, and the driving track is expanded to a certain width on the left and right, as a safe area for the mining truck to travel. Calculate the distance from each convex hull point to the driving track, and judge whether the cluster is in the safe detection area of the mining card according to the distance value and the width of the safe area. As long as the distance value of one convex hull point is less than the width of the safe area, the obstacle is considered to be in the safe area, otherwise it is not in the safe area.

The fourth step is to check whether the vehicle is in the drivable area. As shown in Figure 3: In order to ensure the safe driving of mining trucks, it is also necessary to effectively detect the edge of the road to prevent the mining trucks from driving into the space outside the road when the positioning information is invalid.

The drivable area detection method based on the area growing algorithm is adopted. For the 0 and 1 binary map of "ground-elevated point", select the center point in front of the mine card as the seed point, and use the four-neighborhood growth method to extract the non-obstacle area in front of the mine card, and the driving area can be obtained. Calculate the distance value between the center point in front of the mine card and the boundary point of the drivable area, sort the distance value data to obtain the minimum value, and compare the minimum value with the preset threshold. If the minimum value is less than the preset threshold If the threshold is set, the vehicle is in danger of encountering obstacles, otherwise, it is considered that the vehicle will not encounter obstacles in the drivable area.

The fifth step is to match the lidar data with the millimeter-wave radar data based on the matching distance threshold, and output the final result. The advantage of lidar is that it has a wider detection range and higher detection accuracy, but its performance is poor in extreme weather such as rain, snow and fog, while millimeter-wave radar has strong penetrating ability. Therefore, in order to improve the accuracy and safety of system detection , this embodiment also uses millimeter-wave radar data for matching, as follows:

First, the millimeter-wave radar data is converted into the vehicle body coordinate system,

The millimeter-wave radar is connected to the de-noising filter, and the target detected by the millimeter-wave radar is de-noised and tracked. Millimeter-wave radar is not affected by conditions such as rain, snow, and dust. Therefore, by matching the detection results of lidar with the detection results of millimeter-wave radar, the false detection rate is reduced. The matching method is: the millimeter wave radar detection result and the lidar detection result are judged whether to pair according to the distance, and the distance threshold is dynamically allocated by the obstacle size information detected by the lidar. If the matching is successful, the lidar detection result is saved as the final result. In the detection result, the distance, size, and orientation information of the obstacles are retained. If the matching fails, the failed matching of the lidar obstacles is filtered out, and the successful matching is retained, and then the effective obstacle information that is successfully matched is output.

In terms of hardware of the present invention, one 32-line laser radar, three millimeter-wave radars, and one inertial navigation system can be selected.

Compared with the prior art, aiming at the actual application environment of mining dump trucks, the present invention proposes an obstacle detection system for self-driving mining dump trucks, which can effectively detect obstacles in the road, prevent missed detection, and accurately. Clustering; effective detection of road edges to ensure safety, especially suitable for unstructured road environments such as mines. And it is also applicable under harsh conditions such as rain and snow, strong and weak light, and dust, and has good robustness. From the perspective of applicable scenarios, the detection target includes structured roads in addition to the mine scene (unstructured road), to prevent the mining card from driving into the space other than the road when the positioning information is invalid.

The present invention has good robustness. According to the working principle and advantages and disadvantages of the laser radar, the present invention also adopts a variety of radar fusion schemes. false detection rate. If the matching is successful, the lidar detection result is saved as the final detection result, and the distance, size, orientation and other information of the obstacles are retained. If the matching fails, the matching failures in the lidar obstacles are filtered out, and the matching is successful, and then the matching Successful output of valid obstacle information.

The above are only the preferred embodiments of the present invention. It should be pointed out that for those skilled in the art, without departing from the technical principle of the present invention, several improvements and modifications can also be made. These improvements and modifications It should also be regarded as the protection scope of the present invention.

Claims (10)

1. A method for detecting obstacles of an unmanned mine card is characterized by comprising the following steps: the method comprises the following steps:

respectively converting barrier data acquired by a laser radar and a millimeter wave radar into corresponding vehicle body coordinate systems;

drawing a 0 and 1 binary image of a ground-elevated point by combining the height difference of the grid map and the ground detection of the domain difference value;

clustering elevated points by adopting a multi-parameter model;

judging whether the clustering result is an obstacle influencing the normal running of the vehicle or not according to the motion track of the vehicle;

detecting whether the vehicle is in a drivable area;

and matching the obstacle data acquired by the millimeter wave radar with the obstacle data acquired by the laser radar, and outputting a final result.

2. The unmanned mine card obstacle detection method according to claim 1, wherein: the method for drawing the 0 and 1 binary image of the ground-elevated point by combining the height difference of the grid map and the ground detection of the domain difference value comprises the following steps: the method comprises the steps of rasterizing barrier data, obtaining a grid diagram of N x N, calculating a difference value between the maximum value and the minimum value of point cloud elevation in each grid, carrying out gradient calculation on the difference value of each grid, comparing the difference value with a preset threshold value, identifying points with the difference value larger than the threshold value as 'elevated points', carrying out neighborhood difference calculation on point clouds with the difference value smaller than the threshold value, separating the 'elevated points' grid and ground grids, and fitting all ground grids into a road surface.

3. The unmanned mine card obstacle detection method according to claim 2, wherein: the selection of the domain in the neighborhood difference calculation is to identify grids within 2 meters around the grid as the neighborhood.

4. The unmanned mine card obstacle detection method according to claim 2, wherein: the method for separating the ground grid comprises the following steps: dividing all point clouds belonging to the ground into a plurality of point cloud intervals according to the distribution of the vehicle distances, performing plane fitting on the point cloud intervals, and connecting all fitted planes to obtain the road surface.

5. The unmanned mine card obstacle detection method according to claim 1, wherein: and clustering the elevated points by using the multi-parameter model by adopting an eight-connected domain method to obtain elevated point cluster clusters, and merging a plurality of elevated point cluster clusters into one according to merging parameters to finish the target extraction of the point cloud.

6. The unmanned mine card obstacle detection method of claim 5, wherein: the merging parameters comprise the size and the density of the overhead point cluster.

7. The unmanned mine card obstacle detection method according to claim 1, wherein: and converting the clustered elevated point cluster into two-dimensional data, calculating convex hull information of the two-dimensional data, and judging whether the safe driving of the vehicle is influenced or not according to the convex hull information and the driving track of the vehicle.

8. The unmanned mine card obstacle detection method of claim 7, wherein: and the left and right sides of the driving track of the vehicle are respectively extended by a certain width to be used as a safety zone of the vehicle, and if the safety zone has the elevated point cluster, the obstacle is determined to exist in the safety zone.

9. The unmanned mine card obstacle detection method according to claim 1, wherein: the method for detecting whether the vehicle is in the drivable region comprises the following steps: selecting a certain point in front of the vehicle as a seed point, adopting a four-neighborhood growing mode, reserving a non-obstacle area in front of the mine card, calculating a distance value between a center point in front of the vehicle and the non-obstacle area, sequencing the distance value data to obtain a minimum value, comparing the minimum value with a set threshold value, if the minimum value is smaller than the set threshold value, adjusting the vehicle, otherwise, determining that the vehicle is safe.

10. The unmanned mine card obstacle detection method according to claim 9, wherein: and selecting the center point in front of the vehicle as a seed point.

Images