Obstacle recognition method and device based on single-line laser radar with tilt installation

Abstract

The application discloses a kind of obstacle identification method and device based on single-line laser radar of inclination installation.The identification method includes: the point cloud data obtained by single-line laser radar of inclination installation is preprocessed;The point cloud data after pre-processing is screened, remove the point cloud data corresponding to ground point and slope point therein, obtain the point cloud data after screening;The clustering analysis is carried out to the point cloud data after screening;The clustering data with same label are merged;Wherein, the removal is determined according to the calculation of the point cloud data PC t+1 of current frame and the point cloud data PC t of its previous frame.The application can effectively remove ground and slope and other non-obstacles in identification process, without additional auxiliary equipment, low in cost and efficient.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of robot obstacle recognition methods, in particular to the technical field of obstacle recognition methods based on single-line laser radars. BACKGROUND

[0002] On robots such as unmanned vehicles, single-line laser radars are usually installed horizontally to avoid the problem of obstacle mis-detection caused by laser hitting the ground. The horizontal installation of single-line laser radars usually includes the following installation methods: (1) installing the single-line laser radar on the bottom of the vehicle head and tail on both sides, close to the ground. Although this installation method can avoid the problem of mis-detection caused by laser hitting the ground, it is easy to cause the laser radar to be dirty, thereby causing the unmanned vehicle to misjudge that there is an obstacle in front due to laser hitting the dirty laser radar. In addition, the laser radar in this installation method is prone to damage due to collision; (2) installing the single-line laser radar on the vehicle head and tail on both sides, above the vehicle bottom but close to the vehicle bottom. Compared with the installation method (1), the installation height is higher. This installation method not only avoids the problem of mis-detection caused by laser hitting the ground, but also avoids the problem of laser radar being easy to be dirty and damaged due to installation at the bottom. However, due to the high installation height, obstacles lower than the installation height of the laser radar are difficult to be scanned by the single-line laser radar sensor, resulting in missed detection and safety hazards.

[0003] To solve the above technical problems, a feasible technical means is to improve the single-line laser radar from horizontal installation to inclined installation, such as adjusting the installation method to be installed on the vehicle head and tail on both sides, above the vehicle bottom but close to the vehicle bottom, and inclined downward. This installation method has the advantages of the above method (2), and because there is an inclination angle between the laser radar emitting surface and the ground, the laser can scan the ground and detect low obstacles below the installation position within a certain range, avoiding missed detection and ensuring the safety of the unmanned vehicle.

[0004] However, the above three installation methods all have a common defect, that is, the single-line laser radars using the above three installation methods will misjudge the slope and uneven ground as obstacles when the unmanned vehicle runs on uneven ground, especially uphill and downhill, thereby causing decision errors of the unmanned vehicle.

[0005] To overcome the above defects, some existing improvement methods are as follows: additional equipment is installed on the single-line laser radar, so that the radar can realize up and down scanning, thereby obtaining the three-dimensional coordinates of the scanned object and producing an effect similar to that of a multi-line laser radar. However, because additional equipment is installed, the overall structure becomes more complex, the cost increases, and the reliability decreases. SUMMARY

[0006] In view of the defects of the prior art, the purpose of the present application is to provide an obstacle identification method and device based on a single-line laser radar in a tilted installation mode, which can remove non-obstacles such as the ground and slopes in the identification process, protect the safe operation of an unmanned vehicle without using additional auxiliary equipment, and is low in cost and simple and efficient in calculation process.

[0007] The technical solutions of the present application are as follows:

[0008] The obstacle identification method based on the single-line laser radar in the tilted installation mode comprises:

[0009] S3, pre-processing point cloud data obtained by the single-line laser radar in the tilted installation mode to obtain pre-processed point cloud data;

[0010] S4, screening the pre-processed point cloud data to remove point cloud data corresponding to ground points and slope points, to obtain screened point cloud data;

[0011] S5, performing cluster analysis on the screened point cloud data to obtain cluster data;

[0012] S6, merging cluster data with the same label, and removing point cloud data with a point number less than a point number threshold thresh p after the merging to obtain an identification result;

[0013] Wherein, the removing is determined according to the variance of the point cloud data PC t+1 at the current frame, i.e., at time t+1, and the point cloud data PC t at the previous frame, i.e., at time t, and the slope between extraction points randomly extracted from the point cloud data PC t at time t and corresponding points of the extraction points in the point cloud data PC t+1 at time t+1.

[0014] According to some specific embodiments of the present application, the removing comprises:

[0015] Setting a height threshold h, and removing point cloud data with a height data greater than or equal to the height threshold h, i.e., removing point cloud data with x , wherein x represents the abscissa of any point in the point cloud data PC t+1 , and y represents the ordinate of the point;

[0016] Setting a distance threshold d, and removing point cloud data with a distance data greater than or equal to the distance threshold d, i.e., removing point cloud data with x .

[0017] According to the obstacle identification method according to claim 2, the removing further comprises:

[0018] Calculate the point cloud data PC at time t t Variance of coordinate values ​​in the x-direction variance of coordinates in the y-direction and point cloud data PC at time t+1 t+1 Variance of coordinate values ​​in the x-direction variance of coordinates in the y-direction

[0019] Point cloud data PC at random time t t Data from N points is extracted from the PC. t If the number of points is less than the set extraction number N, then all the data of those points are extracted to obtain the extraction set P at time t, as follows: Where i represents the starting extraction point number;

[0020] PC for obtaining point cloud data at time t+1 t+1 The N corresponding points whose y-coordinates differ from those in the extraction set P by ±yd are used to form the extraction set Q at time t+1, as follows:

[0021] When PC t+1 There exist multiple ±y values ​​for any point in P. d When selecting points corresponding to the range, choose |y d The smallest point is taken as the point in Q corresponding to any point in P;

[0022] Calculate the slope s of the line connecting a point in the extraction set P at time t to its corresponding point in the extraction set Q at time t+1, and obtain the slope set S as follows: S={s i s i+1 s i+2 s i+3 ......s i+N} Calculate the mean slope AVE of the slope set S. s And obtain the slope angle corresponding to the mean slope, i.e., the mean slope angle θ. s ;

[0023] Set coordinate value threshold thresh var and tilt angle threshold thresh s And make the following judgment:

[0024] When |θ s | Less than the tilt angle threshold thresh s ,and and Less than the coordinate value threshold thresh var If the data is not clear, then the point cloud data at this moment is considered to be ground point cloud data and needs to be removed.

[0025] when |θ s is greater than the inclination angle threshold thresh s , and and is less than the coordinate value threshold thresh var , the point cloud data at this moment is considered to be slope point cloud data, which needs to be removed;

[0026] If and is greater than the coordinate value threshold thresh var , the point cloud data at this moment is considered to be uneven or obstacle point cloud data, which needs to be retained.

[0027] According to some embodiments of the present application, the line slope s of the point pair is obtained by the following calculation method:

[0028]

[0029] wherein x1, y1 are the horizontal and vertical coordinates of any extraction point p(x1, y1) in the extraction set P at time t, x2, y2 are the horizontal and vertical coordinates of the corresponding point q(x2, y2) of the extraction point in the extraction set Q at time t+1, v is the speed of the movable carrier on which the single-line laser radar is installed at an inclination, θ is the installation inclination angle of the single-line laser radar on the movable carrier, and k is the time interval between time t and time t+1.

[0030] According to some embodiments of the present application, the preprocessing includes denoising and / or obtaining a region of interest.

[0031] According to some embodiments of the present application, the obstacle recognition method further comprises:

[0032] S1 obtaining the current frame point cloud data of the single-line laser radar, i.e. the point cloud data at time t+1, and determining whether the point cloud data is empty and whether there is non-empty point cloud data at the previous frame, i.e. at time t;

[0033] S2 if there is no non-empty point cloud data at the previous frame and the current frame point cloud data is non-empty, saving the current frame point cloud data and returning to step S1; if there is no non-empty point cloud data at the previous frame and the current frame point cloud data is empty, directly returning to step S1; if there is non-empty point cloud data at the previous frame and the current frame point cloud data is empty, directly returning to step S1; if there is non-empty point cloud data at the previous frame and the current frame point cloud data is non-empty, proceeding to step S3.

[0034] According to some embodiments of the present application, the cluster analysis uses one or more of the Euclidean cluster method and the DBSCAN cluster method.

[0035] The application further discloses an obstacle identification device, which comprises a single-line laser radar obliquely installed on a carrier and a computing device for implementing the obstacle identification method.

[0036] According to some specific embodiments of the application, the single-line laser radar is obliquely installed downward above and close to the bottom of the head and / or tail of the carrier.

[0037] According to some specific embodiments of the application, the computing device comprises a processor and a memory, the memory stores an executable program executable by the processor, and the executable program can complete the obstacle identification method when executed.

[0038] The identification method of the application can distinguish non-obstacle conditions with different slopes, such as a slope with a large slope and a flat ground with a slope close to zero. In addition, considering that it is difficult to have a slope actually equal to zero in reality, different thresholds are introduced in the calculation, and the accuracy of identification can be significantly improved by setting the thresholds.

[0039] In the judgment of the obstacle body, the application effectively utilizes the situation that the point cloud fluctuation state between different frames is related to the obstacle shape, and the accurate obstacle identification effect is obtained by the variance change of the coordinate values in the point cloud.

[0040] The application has the following beneficial effects:

[0041] Compared with a general multi-line laser radar, the application can obtain effective near-distance point cloud data and realize accurate obstacle avoidance directly through the oblique installation of the single-line laser radar without additional equipment, thereby reducing the equipment cost and increasing the reliability of the equipment.

[0042] The application creatively proposes to use the upper and lower frame point cloud data to judge the ground, slope and other non-obstacles, solves the problem of difficulty in judging the ground and slope in the prior art, and can be realized only by simple upper and lower frame point cloud calculation, such as point cloud slope and point cloud variance, thereby greatly improving the judgment efficiency and accuracy, without the need of additional equipment, improving the work efficiency and saving the work cost. BRIEF DESCRIPTION OF DRAWINGS

[0043] Figure 1 It is a schematic diagram of the oblique installation of the single-line laser radar of the application.

[0044] Figure 2 It is a flow chart of the obstacle identification in the specific embodiments.

[0045] Figure 3 Fig. 1 is a schematic diagram of a ground scanning model of a single-line laser radar in the embodiment.

[0046] Figure 4 Fig. 2 is a schematic diagram of a driving model of an unmanned vehicle in the embodiment.

[0047] Figure 5 Fig. 3 is a schematic diagram of a slope calculation method in the embodiment. EMBODIMENT

[0048] The present application is described in detail below in conjunction with the embodiments and drawings, but it should be understood that the embodiments and drawings are only used to exemplarily describe the present application and cannot constitute any limitation on the protection scope of the present application. All reasonable transformations and combinations within the scope of the inventive concept of the present application fall within the protection scope of the present application.

[0049] Referring to the drawings, Figure 1 the installation mode of the single-line laser radar in the present application is that the single-line laser radar 2 is installed obliquely downward above the vehicle bottom near the vehicle bottom on both sides of the vehicle head and / or vehicle tail of the unmanned vehicle 1.

[0050] In this installation mode, referring to the drawings, Figure 2 the present application obtains the recognition result through the following calculation process:

[0051] S1 obtains the current frame point cloud data of the single-line laser radar, i.e., the point cloud data at time t+1, and judges whether the point cloud data is empty and whether there is non-empty point cloud data at the previous frame, i.e., at time t;

[0052] S2 if there is no non-empty point cloud data at the previous frame and the current frame point cloud data is non-empty, the current frame point cloud data is saved and returned to step S1; if there is no non-empty point cloud data at the previous frame and the current frame point cloud data is empty, it is directly returned to step S1; if there is non-empty point cloud data at the previous frame and the current frame point cloud data is empty, it is directly returned to step S1; if there is non-empty point cloud data at the previous frame and the current frame point cloud data is non-empty, step S3 is performed;

[0053] S3 pre-processes the point cloud data of the non-empty current frame and the previous frame, such as removing noise, obtaining the point cloud of the region of interest, etc.;

[0054] S4 screens the pre-processed point cloud data to remove the ground point data therein, to obtain the screened point cloud data;

[0055] S5 performs clustering analysis on the screened point cloud data to obtain the clustering data;

[0056] S6 merges cluster data of the same object with different labels and removes thresh data where the number of points after merging is less than the threshold. p The point cloud data is used to obtain the output results.

[0057] Furthermore, obtaining the filtered point cloud data includes:

[0058] S41 (Refer to Appendix) Figure 3 Establish a ground scanning model for a single-line lidar, where the coordinates of the contact point between the lidar's scanning ray and the ground are (x, y), and the tilt angle of the ray relative to the ground is θ. Then, the distance from the lidar to the contact point (x, y) is: The height of the laser radar reaching the ground is:

[0059] S42, based on the ground scanning model of the single-line lidar, removes the ground data from the preprocessed point cloud data, specifically including:

[0060] S421 sets a height threshold h, removing point cloud data with heights greater than or equal to this height threshold h, i.e., removing... The point cloud data; this height threshold can be set according to the installation height of the lidar and the minimum height to be detected;

[0061] S422 sets a distance threshold d, and removes point cloud data with a distance greater than or equal to the distance threshold d, i.e., removes... The point cloud data; this distance threshold can be set based on experience;

[0062] S423 is established as follows (see attached) Figure 4 The autonomous vehicle driving model shown depicts vehicle 1 driving within the road boundaries. Due to the small time interval between frames and the structural limitations of the vehicle, it only undergoes translation between time t and t+1, without rotation. Furthermore, the translation primarily occurs in the x-direction; that is, its radar coordinate system only undergoes translation in the x-direction between time t and t+1. Its point cloud data at time t is PC. t Point cloud data at time t+1 is from PC t+1 The coordinates in the x-direction are Px t Px t+1 The coordinates in the y-direction are Py t Py t+1 ;

[0063] S424 in the driving model of S423:

[0064] (1) Calculate the point cloud data PC at time t. t Variance of coordinate values ​​in the x-direction and y direction coordinate value variance point cloud data PC at t+1 time t+1 x direction coordinate value variance and y direction coordinate value variance The variance calculation formula is: Wherein x is the sample value, μ is the sample mean, and n is the total number of samples;

[0065] (2) Randomly extract N points of data in the point cloud data PC at t time t When the number of points of PC t is less than the set extraction number N, the data of all points is extracted, and the extraction set P at t time is obtained, as follows: Wherein, i represents the starting extraction point serial number;

[0066] (3) Obtain N corresponding points in the point cloud data PC t+1 at t+1 time, which are the difference between the coordinate values of each point in the extraction set P and the y direction, ±y d , to form the extraction set Q at t+1 time, as follows: When there are multiple points in PC t+1 corresponding to the ±y d range of any point in P, select the point with the smallest |y d | as the point in Q corresponding to the point in P, wherein y d may be set according to empirical values;

[0067] (4) Calculate the slope of the point pair formed by the points in the extraction set P at t time and their corresponding points in the extraction set Q at t+1 time, obtain the slope set S, as follows: S = {s i , s i+1 , s i+2 , s i+3 ........s i+N}, and calculate the slope mean AVE s of the slope set S and the inclination angle corresponding to the slope mean, i.e. the mean inclination angle θ s ;

[0068] (5) Set the coordinate value threshold thresh var and the inclination angle threshold thresh s , and make the following judgments:

[0069] When |θ s | is less than the inclination angle threshold thresh s , and and are both less than the coordinate value threshold thresh varIf the point cloud data at this moment is considered to be ground point cloud data, it needs to be removed.

[0070] When |θ s | is greater than the inclination angle threshold thresh s , and and are less than the coordinate value threshold thresh var , the point cloud data at this moment is considered to be slope point cloud data, which needs to be removed.

[0071] If and are greater than the coordinate value threshold thresh var , the point cloud data at this moment is considered to be uneven or obstacle point cloud data, which needs to be retained.

[0072] Further, with reference to the accompanying drawings, Figure 5 (wherein, DH, FI are laser lines, FG, DE are the positions of the vehicle at time t and t+1), the slope of the connecting line of the point pair is calculated as follows:

[0073] It is assumed that the single-line laser radar advances on the horizontal ground and scans the slope SL at time t and t+1, the point cloud data of the scanning point on the slope in the extraction set P at time t is p(x1, y1), the point cloud data of the scanning point in the extraction set Q at time t+1 is q(x2, y2), the vehicle speed is v, the inclination angle of the single-line laser radar is θ, and the time interval between time t and t+1, i.e., the time interval between the upper and lower frames, is k. Then, the slope s is calculated as follows:

[0074] ,

[0075] The corresponding slope angle ∠HIJ = arctan(s).

[0076] When the ground is scanned, the slope is calculated in the same way.

[0077] Further, the threshold thresh var and the inclination angle threshold thresh s may be set according to the installation inclination angle of the vehicle body laser radar, the point cloud quality of the laser radar, and the experience value.

[0078] Further, the clustering analysis in step S5 can use any one or more of the clustering methods such as Euclidean clustering, density-based spatial clustering of applications with noise (DBSCAN), and the like.

[0079] Further, the point threshold thresh in step S6 p The point threshold thresh can be set according to the installation tilt angle of the vehicle body laser radar, the point cloud quality of the laser radar, and an experience value.

[0080] The above embodiments are only preferred embodiments of the present application, and the protection scope of the present application is not limited to the above embodiments. Any technical scheme falling within the concept of the present application belongs to the protection scope of the present application. It should be pointed out that improvements and refinements made by those skilled in the art without departing from the principles of the present application should also be considered as falling within the protection scope of the present application.

Claims

1. A method for obstacle recognition based on a tilt-mounted single-line laser radar, characterized by, It comprises: S3 pre-processes the point cloud data obtained by the single-line laser radar installed obliquely to obtain pre-processed point cloud data; S4 filters the pre-processed point cloud data to remove point cloud data corresponding to ground points and slope points, to obtain filtered point cloud data; S5 performs cluster analysis on the filtered point cloud data to obtain cluster data; S6 merging the clustered data with the same label, and removing the merged points less than the point threshold point cloud data, obtaining a recognition result; Wherein, the removing is according to the point cloud data of the current frame, i.e. t+1 time and the point cloud data of the previous frame, i.e. t time In the case of variance in x, y two directions, and the extracted points obtained from the point cloud data of t time and the corresponding points of these extracted points in the point cloud data of t+1 time Determination of the slope between the extracted points The removal comprises: A height threshold h is set, and the point cloud data whose height data is greater than or equal to the height threshold h is removed, that is, the point cloud data of is removed, wherein x represents the horizontal coordinate of any point in the point cloud data , and y represents the vertical coordinate of the point. A distance threshold d is set, and the point cloud data whose distance data is greater than or equal to the distance threshold d is removed, that is, the point cloud data of is removed. Point cloud data at time t Coordinate value variance in the x direction Coordinate value variance in the y direction Point cloud data at time t+1 Coordinate value variance in the x direction Coordinate value variance in the y direction ; Random point cloud data at time t Extract N points of data, when The number of points is less than the set extraction number N, then extract all point data, get the extraction set P at time t, as follows: Where i represents the starting extraction point sequence number; acquire point cloud data at time t+1 the difference between the coordinate value in the y direction of each point in the extraction set P and N corresponding points of the point cloud data at time t, form an extraction set Q at time t+1, as follows: When there are multiple points in Q corresponding to the range of any point in P, the point with the smallest value is selected as the point in Q corresponding to the point in P. When there are multiple points in Q corresponding to the range of any point in P, the point with the smallest value is selected as the point in Q corresponding to the point in P. The slope s of the line connecting the point pair formed by the point in the extraction set P at the t moment and its corresponding point in the extraction set Q at the t+1 moment is calculated to obtain a slope set S, as follows: The average value of the slopes of the slope set S is calculated and the inclination angle corresponding to the average value of the slopes, i.e. the average inclination angle is obtained; Setting a coordinate value threshold and a tilt angle threshold and making the following determination: When less than the tilt angle threshold , and and are less than the coordinate threshold , the point cloud data at this moment is considered as ground point cloud data, which needs to be removed. When greater than the inclination angle threshold , and and are both less than the coordinate threshold , the point cloud data at this moment is considered to be slope point cloud data and needs to be removed. If and are greater than the coordinate threshold , it is considered that the point cloud data at this moment is the point cloud data of the concave-convex place or obstacle, which needs to be retained.

2. The obstacle recognition method according to claim 1, characterized by, The slope s of the connecting line of the point pair is obtained by the following calculation method: ， wherein x1, y1 are the horizontal and vertical coordinates of any sampling point in the sampling set P at the time t x2, y2 are the horizontal and vertical coordinates of the corresponding point in the sampling set Q at the time t+1 v is the speed of the movable carrier on which the single-line laser radar is installed obliquely, is the installation oblique angle of the single-line laser radar on the movable carrier, and k is the time interval between the time t and the time t+1.

3. The obstacle recognition method according to claim 1, characterized by, The pre-processing comprises denoising and / or obtaining a region of interest.

4. The obstacle recognition method according to claim 1, characterized by, It further comprises: S1 obtains the current frame point cloud data of the single-line laser radar, i.e. the point cloud data at time t+1, and judges whether the point cloud data is empty and whether there is non-empty point cloud data at the previous frame, i.e. at time t; S2 if there is no non-empty point cloud data at the previous frame, and the current frame point cloud data is non-empty, save the current frame point cloud data and return to step S1; if there is no non-empty point cloud data at the previous frame, and the current frame point cloud data is empty, directly return to step S1; if there is non-empty point cloud data at the previous frame, and the current frame point cloud data is empty, directly return to step S1; if there is non-empty point cloud data at the previous frame, and the current frame point cloud data is non-empty, proceed to step S3.

5. The obstacle recognition method according to claim 1, characterized by, The cluster analysis uses one or more of the Euclidean cluster method and the DBSCAN cluster method.

6. An obstacle recognition device comprising a single-line laser radar obliquely installed on a carrier and a computing device implementing the obstacle recognition method of any one of claims 1-5.

7. The obstacle recognition device according to claim 6, characterized in that The single-line laser radar is installed obliquely downward above the bottom and close to the bottom on both sides of the head and / or tail of the carrier.

8. The obstacle recognition device according to claim 6, characterized in that The computing device comprises a processor and a memory, the memory storing an executable program executable by the processor, the executable program being executable to complete the obstacle recognition method when executed.

Images