2d laser point cloud detection method for right-angle corner point of greenhouse intersection and storage medium

Abstract

The application belongs to the technical field of point cloud recognition, and specifically relates to a 2D laser point cloud detection method for right-angle corner points of a greenhouse intersection and a storage medium. The method comprises the following steps: acquiring a single-frame laser point cloud of a robot at a certain moment; establishing a cross sliding window; performing sliding detection on all points in the single-frame laser point cloud by the cross sliding window, so as to detect right-angle corner points; constructing a point cloud set of the right-angle corner points according to the detected right-angle corner points; and calculating optimal right-angle corner points in the point cloud set. The application can effectively extract right-angle corner points from noise data, and has better timeliness than an image Harris corner point detection algorithm. The application patent consumes less than half of the time of the image method.

Description

Technical Field

[0001] This invention belongs to the field of point cloud recognition technology, specifically a 2D laser point cloud detection method and storage medium for right-angle corner points at greenhouse intersections. Background Technology

[0002] In intersections and passageways in locations such as greenhouses, right-angled corners are common. Identifying right-angled corners in a single frame of 2D laser point cloud is crucial for estimating the real-time position of robots at intersections. For corner feature extraction, the common approach is to first extract line segment features, then calculate the intersection of two line segments to obtain the corner features. Another method is image-based, which requires converting the point cloud into an image, then using the Harris corner detection algorithm to identify corners in the image, and finally converting the pixel coordinates of the corners into actual coordinates.

[0003] Existing methods require cumbersome steps, cannot directly extract right-angle corner points from the point cloud, and have low extraction efficiency. Existing methods are affected by the accuracy of line fitting or image resolution, and are also affected by noise from the LiDAR sensor, resulting in low accuracy in right-angle corner point identification. Summary of the Invention

[0004] To address the aforementioned problems, this invention provides a 2D laser point cloud detection method and storage medium for right-angled corner points at greenhouse intersections.

[0005] This invention adopts the following technical solution: a 2D laser point cloud detection method for right-angle corner points at greenhouse intersections, comprising:

[0006] Obtain a single frame of laser point cloud in 2D model of the robot at a certain moment;

[0007] The cross-shaped sliding window was constructed;

[0008] A cross-shaped sliding window performs sliding detection on all points in a single frame of laser point cloud to detect right-angle points;

[0009] Construct a point cloud set of right-angled corner points based on the detected right-angled corner points;

[0010] The optimal right-angled corner point is calculated from the point cloud set.

[0011] In some embodiments, the acquired single-frame laser point cloud needs to be rotated counterclockwise. , This represents the robot's orientation angle.

[0012] In some embodiments, the cross-shaped sliding window is divided into 5 zones, from 0 to 4;

[0013] Region 0 is the corner region, which is a square and is set at the very center.

[0014] Regions 1-4 are channel edge regions, which are rectangular and are set to the right, top, left, and bottom of region 0, respectively.

[0015] In some embodiments, the horizontal side length of region 2,0,4 and the vertical side length of region 1,0,3 are the same, both being 2l2.

[0016] In some embodiments, , The standard deviation of the point cloud data is given.

[0017] In some embodiments, half of the long side of the cross-shaped sliding window is l1, and l1 is less than the channel width. w .

[0018] In some embodiments, the sliding detection includes:

[0019] The detection process starts from the first point and ends at the last point.

[0020] When the right-angle point is located at the lower left corner, lower right corner, or upper right corner of the right-angled wall, if the following conditions are met... If so, it is considered that a right-angled corner point has been detected;

[0021] When the right-angle point is located at the upper left corner of the right-angled wall, if the following conditions are met... If so, it is considered that a right-angled corner point has been detected;

[0022] in, The area code for points outside of zone 0. and These represent the operations to extract the maximum and minimum values, respectively. Indicates the number of data items in the set. This represents the set of data belonging to region 1, where the index j indicates the set it belongs to. .

[0023] In some embodiments, if the number of point clouds in a certain zone from 1 to 4 is less than a set threshold, the statistical value of the number of point clouds in that zone is set to 0.

[0024] In some embodiments, calculating the optimal right-angle corner point includes:

[0025] When the right-angled corner point is located at the lower left corner of the right-angled wall, establish the objective function for minimizing the mixed distance:

[0026]

[0027] in The objective function for this situation; This represents the number of elements in zones 0 and 2 excluding point i, moving counterclockwise from point i. This represents the number of elements in zones 0 and 1 excluding point i, moving clockwise from point i.

[0028] The right-angled corner point is located at the lower right corner of the right-angled wall. Establish the objective function to minimize the mixed distance:

[0029]

[0030] in The objective function for this situation; This represents the number of elements in zones 0 and 3 excluding point i, moving counterclockwise from point i. This represents the number of elements in zones 0 and 2 excluding point i, moving clockwise from point i.

[0031] The right-angled corner point is located at the upper right corner of the right-angled wall. Establish the objective function to minimize the mixed distance:

[0032]

[0033] in The objective function for this situation; This represents the number of elements in zones 0 and 4 excluding point i, moving counterclockwise from point i. This represents the number of elements in zones 0 and 3 excluding point i, moving clockwise from point i.

[0034] When the right-angled corner point is located at the top left corner of the right-angled wall, establish the objective function for minimizing the mixed distance:

[0035]

[0036] in The objective function for this situation; This represents the number of elements in zones 0 and 1 excluding point i, moving counterclockwise from point i. This represents the number of elements in zones 0 and 4 excluding point i, moving clockwise from point i.

[0037] The optimal right-angle point is obtained by solving the objective function for the minimum value.

[0038] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a 2D laser point cloud detection method for right-angled corner points at greenhouse intersections.

[0039] Compared with the prior art, the present invention has the following beneficial effects:

[0040] This invention directly processes 2D single-frame laser point clouds, including: rotating the laser radar point data based on the robot's orientation angle; establishing a cross-shaped sliding window region based on sensor noise levels and channel width, and establishing right-angle corner detection criteria for different cases, sequentially sliding the detection from the first point to the last point to obtain the point cloud falling within the corner region; establishing an optimal right-angle corner detection objective function, thereby identifying the optimal right-angle corner from the point cloud of the corner region. The right-angle corners identified by this invention can provide feature reference points for real-time position estimation of the robot at intersections.

[0041] This invention proposes a 2D single-frame laser point cloud cross-window detection algorithm for right-angle corner detection at intersections. It is more robust than the image Harris corner detection algorithm and can effectively extract right-angle corners from noisy data. Furthermore, it is more efficient than the image Harris corner detection algorithm, with the time required by this invention being less than half that of the image method. Attached Figure Description

[0042] Figure 1 A schematic diagram illustrating the application of the right-angle corner detection algorithm for cross-shaped sliding windows at intersections;

[0043] Figure 2 The cross-shaped sliding window area is designed to account for sensor noise.

[0044] Figure 3 For different types of right-angle corner points and their detection;

[0045] Figure 4 For the identification of the optimal right-angle corner point;

[0046] Figure 5 A simulation of a single-frame 2D point cloud for right-angle corner recognition at a crossroads;

[0047] Figure 6 A simulation of a single-frame 2D point cloud for right-angle corner recognition at a crossroads;

[0048] Figure 7 This is a right-angle corner point recognition image of a single frame 2D point cloud from an experiment at a crossroads.

[0049] Figure 8 This is a right-angle corner point recognition image of a single frame 2D point cloud from an experiment at a crossroads.

[0050] Figure 9 This is a right-angle corner point recognition image of a single frame 2D point cloud from a T-shaped intersection experiment.

[0051] Figure 10 This is a right-angle corner point recognition image of a single frame 2D point cloud from a T-shaped intersection experiment.

[0052] Figure 11 This is a right-angle corner point recognition image of a single frame 2D point cloud from an L-shaped intersection experiment.

[0053] Figure 12 This is a right-angle corner point recognition image of a single frame 2D point cloud from an L-shaped intersection experiment. Detailed Implementation

[0054] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0055] A 2D laser point cloud detection method for right-angle corner points at temperature intersections, comprising:

[0056] S1: Obtain a single frame of laser point cloud of the robot at a certain moment in 2D;

[0057] S2: The created cross-shaped sliding window;

[0058] S3: The cross-shaped sliding window performs sliding detection on all points in a single frame of laser point cloud, used to detect right-angle points;

[0059] S4: Construct a point cloud set of right-angled corner points based on the detected right-angled corner points;

[0060] S5: Calculate the optimal right-angled corner point in the point cloud set.

[0061] In a specific embodiment, step S1 specifically includes:

[0062] like Figure 1 As shown, a Cartesian coordinate system The robot's orientation angle is known to be fixed at the center of the robot's lidar. The lidar point data is represented in this rectangular coordinate system and rotated counterclockwise. To obtain the representation of the set of n laser point clouds in a single frame at a certain moment: .

[0063] In a specific embodiment, in step S2, the cross sliding window is divided into 5 areas from 0 to 4; where area 0 is the corner area, which is a square and is set in the center; areas 1-4 are the channel edge areas, which are rectangles and are set to the right, top, left and bottom of area 0 respectively.

[0064] First, let's start with points. To establish such as Figure 1 and 2The cross-shaped sliding window shown is divided into five areas, numbered 0 to 4. Area 0 is the corner area, and the others are the passageway edges. The horizontal side lengths of areas 2, 0, and 4 are the same as the vertical side lengths of areas 1, 0, and 3, which are all [missing information]. To improve the robustness of the algorithm to sensor noise, satisfy: It is assumed that the sensor noise is Gaussian white noise. Let be the standard deviation of the point cloud data. Half of the longer side of the cross-shaped sliding window is . , It should be less than the channel width w. The five regions of the cross-shaped sliding window are described by the following inequality:

[0065] .

[0066] In a specific embodiment, in step S3, the cross-shaped sliding window starts detecting from the first point and ends at the last point. The region sequence of the points within the i-th cross-shaped sliding window is as follows:

[0067]

[0068] This represents the center point within the i-th cross-shaped sliding window, i.e. The cross-shaped sliding window area code, .

[0069] Sliding detection includes: starting from the first point and ending at the last point;

[0070] When the right-angle point is located at the lower left corner, lower right corner, or upper right corner of the right-angled wall, if the following conditions are met... If so, it is considered that a right angle has been detected;

[0071] When the right-angle point is located at the upper left corner of the right-angled wall, if the following conditions are met... If so, it is considered that a right-angled corner point has been detected;

[0072] in, The area code for points outside of zone 0. and These represent the operations to extract the maximum and minimum values, respectively. Indicates the number of data items in the set. This represents the set of data belonging to region 1, where the index j indicates the set it belongs to. .

[0073] It should be noted that, in order to filter out the influence of sporadic noise in the channel, if the number of point clouds in a certain zone from 1 to 4 is less than M (this value needs to be adjusted according to actual data), the statistical value of the number of point clouds in that zone is set to 0.

[0074] In a specific embodiment, in step S5, there may be multiple points in the corner region. In order to filter out wall points and obtain the optimal right-angle corner point, a minimum objective function for the mixed distance is established according to the four cases in which the right-angle corner point is located.

[0075] When the right-angled corner point is located at the lower left corner of the right-angled wall, establish the objective function for minimizing the mixed distance:

[0076]

[0077] in The objective function for this situation; This represents the number of elements in zones 0 and 2 excluding point i, moving counterclockwise from point i. This represents the number of elements in zones 0 and 1 excluding point i, moving clockwise from point i.

[0078] When the right-angled corner point is located at the lower right corner of the right-angled wall, establish the objective function for minimizing the mixed distance:

[0079]

[0080] in The objective function for this situation; This represents the number of elements in zones 0 and 3 excluding point i, moving counterclockwise from point i. This represents the number of elements in zones 0 and 2 excluding point i, moving clockwise from point i.

[0081] When the right-angled corner point is located at the upper right corner of the right-angled wall, establish the objective function for minimizing the mixed distance:

[0082]

[0083] in The objective function for this situation; This represents the number of elements in zones 0 and 4 excluding point i, moving counterclockwise from point i. This represents the number of elements in zones 0 and 3 excluding point i, moving clockwise from point i.

[0084] When the right-angled corner point is located at the top left corner of the right-angled wall, establish the objective function for minimizing the mixed distance:

[0085]

[0086] in The objective function for this situation; This represents the number of elements in zones 0 and 1 excluding point i, moving counterclockwise from point i. This represents the number of elements in zones 0 and 4 excluding point i, moving clockwise from point i.

[0087] Finally, the minimum objective function is solved to obtain the optimal right-angle point.

[0088] First, simulation verification was performed, taking the point cloud of a crossroads as an example, such as... Figure 5 and Figure 6 As shown, the simulated point cloud is clean data without noise interference. This invention directly and accurately identified right-angle corner points in four different scenarios, taking 0.2 seconds (test computer configuration: 16-core Intel i9 processor, 128MB RAM, algorithm written in MATLAB 2018b). The image-based method requires first converting the point cloud into an image, then using the Harris corner detection algorithm to identify corner points in the image, and finally converting the pixel coordinates of the corner points into actual coordinates. Due to the influence of image resolution, it cannot accurately identify corner point positions in most cases, and this method also identifies invalid corner points (channel wall edge points), taking 0.45 seconds (under the same test conditions).

[0089] To further illustrate the advantages of this method, we also conducted experimental verification, such as... Figure 7-12 Because point clouds acquired by actual lidar contain noise, image-based methods are quite sensitive to noise and may misidentify noisy points as corner points. This method, however, takes noise into account. Since the corner points identified by this method are points in the actual point cloud, although they do not coincide with the actual corner points due to noise, the identified corner points are still in their vicinity. In this experiment, the average distance between the identified corner points and the actual corner points under various conditions was 0.03m. Therefore, this method can effectively identify right-angle points in a single frame of noisy 2D point cloud data.

[0090] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A 2D laser point cloud detection method for right-angled corner points at greenhouse intersections, characterized in that, include: Obtain a single frame of laser point cloud in 2D model of the robot at a certain moment; Establish a cross-shaped sliding window; The cross-shaped sliding window is divided into five zones, from 0 to 4. Region 0 is the corner region, which is a square and is set at the very center. Regions 1-4 are the channel edge regions, which are rectangular and are set to the right, top, left, and bottom of region 0, respectively. A cross-shaped sliding window performs sliding detection on all points in a single frame of laser point cloud to detect right-angle points; The sliding detection includes: The detection process starts from the first point and ends at the last point. When the right-angle point is located at the lower left corner, lower right corner, or upper right corner of the right-angled wall, if the following conditions are met... If so, it is considered that a right-angled corner point has been detected; When the right-angle point is located at the upper left corner of the right-angled wall, if the following conditions are met... If so, it is considered that a right-angled corner point has been detected; in, The area code for points outside of zone 0. and These represent the operations to extract the maximum and minimum values, respectively. Indicates the number of data items in the set. This represents the set of data belonging to region 1, where the index j indicates the set it belongs to. ; Based on the detected right-angled corner points, construct a point cloud set of right-angled corner points; The optimal right-angled corner point is calculated from the point cloud set.

2. The 2D laser point cloud detection method for right-angled corner points at greenhouse intersections according to claim 1, characterized in that, The acquired single-frame laser point cloud needs to be rotated counterclockwise. , This represents the robot's heading angle.

3. The 2D laser point cloud detection method for right-angled corner points at greenhouse intersections according to claim 1, characterized in that, The horizontal side length of region 2,0,4 is the same as the vertical side length of region 1,0,3, both being 2l².

4. The 2D laser point cloud detection method for right-angled corner points at greenhouse intersections according to claim 3, characterized in that, , The standard deviation of the point cloud data is given.

5. The 2D laser point cloud detection method for right-angled corner points at greenhouse intersections according to claim 1, characterized in that, The length of the cross-shaped sliding window is half of the long side, l1, and l1 is less than the width of the passageway. w .

6. The 2D laser point cloud detection method for right-angled corner points at greenhouse intersections according to claim 1, characterized in that, If the number of point clouds in any of zones 1 to 4 is less than the set threshold, the statistical value of the number of point clouds in that zone is set to 0.

7. The 2D laser point cloud detection method for right-angled corner points at greenhouse intersections according to claim 1, characterized in that, Calculating the optimal right-angle point includes: When the right-angled corner point is located at the lower left corner of the right-angled wall, establish the objective function for minimizing the mixed distance: in The objective function for this situation; This represents the number of elements in zones 0 and 2 excluding point i, moving counterclockwise from point i. This represents the number of elements in zones 0 and 1 excluding point i, moving clockwise from point i. When the right-angled corner point is located at the lower right corner of the right-angled wall, establish the objective function for minimizing the mixed distance: in The objective function for this situation; This represents the number of elements in zones 0 and 3 excluding point i, moving counterclockwise from point i. This represents the number of elements in zones 0 and 2 excluding point i, moving clockwise from point i. When the right-angled corner point is located at the upper right corner of the right-angled wall, establish the objective function for minimizing the mixed distance: in The objective function for this situation; This represents the number of elements in zones 0 and 4 excluding point i, moving counterclockwise from point i. This represents the number of elements in zones 0 and 3 excluding point i, moving clockwise from point i. When the right-angled corner point is located at the top left corner of the right-angled wall, establish the objective function for minimizing the mixed distance: in The objective function for this situation; This represents the number of elements in zones 0 and 1 excluding point i, moving counterclockwise from point i. This represents the number of elements in zones 0 and 4 excluding point i, moving clockwise from point i. The optimal right-angle point is obtained by solving the objective function for the minimum value.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the 2D laser point cloud detection method for right-angled corner points at greenhouse intersections as described in any one of claims 1-7.

Images