SLAM mapping and autonomous navigation method for disc brush type photovoltaic cleaning robot

Abstract

The invention discloses a disc brush type photovoltaic cleaning robot SLAM mapping and autonomous navigation method, and the method comprises the steps: obtaining image data through a depth camera, generating a three-dimensional dense point cloud, and separating a cleaning plane point cloud through an RANSAS algorithm; according to the arrangement mode of components in the roof distributed photovoltaic power station with or without inner holes, an NBEM algorithm and an LALSM algorithm are adopted to separate boundary point clouds; constructing a two-dimensional grid map by adopting an octree algorithm; the method comprises the following steps of: planning a full-coverage cleaning path by adopting a boustrophedon algorithm, and dispersing the cleaning path into a plurality of target points; and planning paths among the target points, and controlling the robot to sequentially traverse the target points to complete cleaning work. By adopting the photovoltaic module cleaning robot, a navigation map with accurate boundary information can be constructed, and a path can be autonomously planned to complete full-coverage cleaning, so that the operation and maintenance safety of the photovoltaic power station cleaning robot is improved, and efficient and intelligent cleaning is realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of autonomous navigation of robots, and particularly relates to a method for SLAM mapping and autonomous navigation of a disk-brush type photovoltaic cleaning robot. Background Art

[0002] When driving a photovoltaic cleaning robot to clean photovoltaic modules relying on SLAM (Simultaneous Localization and Mapping) technology, it is necessary to first construct a navigation map of the cleaning area. Existing SLAM mapping methods for photovoltaic power station cleaning robots include vision-based SLAM, lidar-based SLAM, topology-map-based SLAM, and multi-sensor fusion SLAM, etc. The vision-based SLAM mapping method mainly relies on RGB-D cameras, etc. as the main sensors, and constructs a navigation map by extracting feature points in the environment, with the characteristics of low cost, low computational complexity, and high accuracy.

[0003] In a rooftop distributed photovoltaic power station, the scale of the photovoltaic array is small, and the photovoltaic modules need to be flexibly arranged according to the actual terrain of the roof. Therefore, the arrangement forms of rooftop distributed photovoltaic modules are flexible and diverse. The three-dimensional point cloud map generated by the vision-based SLAM technology has the characteristics of unclear semantics. The constructed point cloud map contains information of the entire three-dimensional space and cannot distinguish semantic information such as the edges of the modules and the cleaning areas of the modules. Therefore, it cannot be directly used for the navigation of the cleaning robot. Summary of the Invention

[0004] Aiming at the deficiency that the existing mapping method based on vision-based SLAM technology cannot construct a navigation map with clear semantic information according to the flexible layout of the arrangement of rooftop distributed photovoltaic modules, the present invention discloses a method for SLAM mapping and autonomous navigation of a disk-brush type photovoltaic cleaning robot, which can accurately and quickly extract boundary point clouds from the three-dimensional point cloud map constructed by the vision-based SLAM technology and construct a navigation map of the cleaning area with complete boundary information, and drive the cleaning robot to autonomously plan a path to complete the cleaning.

[0005] The technical solution of the present invention:

[0006] A method for SLAM mapping and autonomous navigation of a disk-brush type photovoltaic cleaning robot includes the following steps:

[0007] Step 1, the cleaning robot obtains image data through a depth camera;

[0008] Step 2, generate a three-dimensional dense point cloud by using the ORBSLAM2 algorithm, and separate the cleaning plane point cloud by using the RANSAS algorithm;

[0009] Step 3: Based on the cleaning plane point cloud, for the two arrangement methods of components in the rooftop distributed photovoltaic power station with or without inner holes, the NBEM algorithm and the LALSM algorithm are respectively used to separate the boundary point cloud;

[0010] Step 4: Using the boundary point cloud as the input, the octree algorithm is used to construct a two-dimensional grid map;

[0011] Step 5: Convert the two-dimensional grid map into a pixel format picture (usually in png format). After preprocessing the pixel format picture (performing erosion and dilation operations), the boustrophedon algorithm is used to plan a full-coverage cleaning path, and the cleaning path is discretized into several target points;

[0012] Step 6: Based on several target points, the A* and TEB algorithms are respectively used to plan the paths between the target points, and at the same time, the robot is controlled to traverse the target points in turn to complete the cleaning work.

[0013] Furthermore, in Step 1, when using the RANSAS algorithm to separate the cleaning plane point cloud, the specific given parameters are: the minimum number of points for the fitting model is 3, the distance threshold from each point to the fitting plane is 0.1 - 0.3 m, and the number of iterations is 3 - 10 times.

[0014] Furthermore, Step 3 specifically includes the following steps:

[0015] Step 3.1: Determine the type of the arrangement method of the photovoltaic components in the rooftop distributed photovoltaic power station;

[0016] Step 3.2: Using the cleaning plane point cloud as the input, for the arrangement method with inner holes, the NBEM algorithm is used to separate the boundary point cloud of the cleaning plane with both inner and outer boundary information; for the arrangement method without inner holes, the LALSM algorithm is used to separate the boundary point cloud of the cleaning plane with only outer boundary information.

[0017] Furthermore, when the NBEM algorithm separates the boundary point cloud of the cleaning plane with both inner and outer boundary information, it specifically includes:

[0018] First, using the cleaning plane point cloud as the input, the NBME algorithm is used to fit an overall plane P and calculate the normal line l of the overall plane n ;

[0019] Secondly, taking the finite number of points around each point p i in the plane point cloud as the input, the NBME algorithm is used to fit a local plane and calculate the normal line of the local plane which is used as the normal line of each point in the cleaning plane point cloud;

[0020] Finally, for each point p i, calculate the normal vector of this point and the included angle between the normal vector l of the overall plane n . Given an angle threshold θ, the boundary point cloud set P b is:

[0021]

[0022] The NBEM algorithm is simple, with a fast iteration speed and a relatively fast speed for separating point clouds.

[0023] Furthermore, the LALSM algorithm separates the boundary point cloud of the cleaning plane that only contains outer boundary information, specifically including:

[0024] First, project the cleaning plane point cloud onto the XOY plane to obtain the two-dimensional plane point cloud P p ;

[0025] Secondly, according to each point p p of the two-dimensional plane point cloud P i , establish the minimum bounding rectangle R:

[0026]

[0027] Among them, and are the horizontal and vertical coordinate values of point p i respectively; (a 1 , b 1 ), (a 1 , b 2 ), (a 2 , b 2 ), (a 2 , b 1 ) are the four vertices of the minimum bounding rectangle respectively.

[0028] Then, given the resolution res, divide the rectangle R into (res + 1) * (res + 1) small rectangles, and the boundary point cloud P b is:

[0029] del x =(a 2 -a 1 ) / res, del y =(b 2 -b 1 ) / res

[0030]

[0031] In the formula: del x , del y are the step sizes with respect to the x and y axes respectively, They are the abscissa and ordinate of the i-th point respectively; They are the minimum and maximum values of the ordinate of the point cloud in the j-th interval respectively; They are the minimum and maximum values of the abscissa of the point cloud in the j-th interval respectively; P b is the set of boundary point clouds.

[0032] The LALSM algorithm can quickly separate the boundary point cloud, with a speed of dozens of milliseconds.

[0033] Furthermore, step 4 is specifically as follows: Using the boundary point cloud as the input, project each point in the boundary point cloud onto the XOY plane by using the octree algorithm to obtain a two-dimensional grid map with a resolution of 0.01 - 0.05 m.

[0034] Furthermore, step 5 is specifically as follows:

[0035] First, convert the two-dimensional grid map into a pixel format picture and perform binarization processing on the pixel format map;

[0036] Secondly, perform erosion and dilation operations on the binarized map;

[0037] Then, use the boustrophedon algorithm to plan a full-coverage cleaning path on the eroded and dilated map;

[0038] Finally, discretize the cleaning path into a finite number of target points, and use the connection path between the target point sets as the theoretically full-coverage cleaning path.

[0039] Furthermore, step 6 is specifically as follows:

[0040] Extract a point from the target point set in sequence as the next target position that the robot needs to reach, and use the A* and TEB algorithms to plan the global and local paths from the current position to the target position respectively, and calculate the speed control amount according to the TEB algorithm to control the robot to reach the target points in sequence until the robot has reached all the target points and completed the cleaning work.

[0041] Compared with the prior art, the present invention has the following beneficial effects:

[0042] The present invention does not require installing a sensing and positioning device around the photovoltaic array. When the arrangement of the rooftop distributed photovoltaic modules changes, the remote control robot walks around in the area of the photovoltaic modules. At this time, the robot can use the mapping algorithm mentioned to construct a map, realizing the construction of a navigation map with complete boundary information according to the flexible arrangement form of the modules in the rooftop distributed photovoltaic power station, and autonomously planning a full-coverage path to drive the cleaning robot to complete full-coverage cleaning. Brief Description of the Drawings

[0043] Figure 1It is the flow chart of the method of the present invention;

[0044] Figure 2 It is the photo of the prototype in Embodiment 1 of the present invention;

[0045] Figure 3 It is the experimental platform in Embodiment 1 of the present invention;

[0046] Figure 4 It is the three-dimensional dense point cloud map generated by using the ORBSLAM2 algorithm;

[0047] Figure 5 It is the point cloud map of the cleaning plane separated by using the RANSAS algorithm;

[0048] Figure 6 It is the point cloud map of the boundary separated by using the LALSM algorithm;

[0049] Figure 7 It is the two-dimensional grid map generated by using the octree algorithm;

[0050] Figure 8 It is the full coverage path planned by using the boustrophedon algorithm;

[0051] Figure 9 It is the global path and local path planned by using the A* algorithm and the TEB algorithm;

[0052] Figure 10 It is the arrangement method and the corresponding navigation map. Specific implementation mode

[0053] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0054] Embodiment 1

[0055] As Figure 10 shown, (a), (b), (c), and (d) in the figure are the arrangement methods of the photovoltaic array panels on the roof. Among them, the arrangement shown in Figure (d) is called the arrangement with inner holes, and the arrangements shown in Figures (a), (b), and (c) are called the arrangements without inner holes.

[0056] As Figure 1 shown, the present method includes:

[0057] Step 1: Place the cleaning robot as Figure 2 shown on the photovoltaic array panels to be cleaned as Figure 3 shown. Control the cleaning robot to move around the edge of the photovoltaic array through the mobile phone app. During this process, the cleaning robot obtains image data through the depth camera.

[0058] Step 2: Generate as Figure 4The three-dimensional dense point cloud shown. Then, the RANSAC algorithm is used to separate the swept plane point cloud as shown in Figure 5 ; the given parameters are: the minimum number of points for the fitting model is 3, the distance threshold from each point to the fitting plane is 0.2 m, and the number of iterations is 5 times.

[0059] Common methods for separating plane point clouds include region growing method, clustering-based method, deep learning-based method, etc. The point cloud data of the rooftop photovoltaic power station may contain noise (such as dust, vegetation). The RANSAC algorithm is robust to noise and outliers, has simple parameter settings, and fast iteration speed.

[0060] Step 3: Figure 3 The photovoltaic array to be cleaned shown is arranged without an "inner hole", so the LALSM algorithm is used to separate the boundary point cloud map as shown in Figure 6 .

[0061] Step 4: Using the boundary point cloud as shown in Figure 6 as the input, the octree algorithm is used to project each point in the boundary point cloud onto the XOY plane, and a two-dimensional grid map with a resolution of 0.04 m as shown in Figure 7 is obtained.

[0062] Step 5: First, convert the two-dimensional grid map into a pixel format picture and perform binary processing on the pixel format map; second, perform erosion and dilation operations on the binary map; then, use the boustrophedon algorithm to plan the full-coverage cleaning path as shown in Figure 8 ; finally, discretize the cleaning path into a finite number of target points, and use the connection path between the target point sets as the theoretically full-coverage cleaning path.

[0063] Step 6: Extract a point from the target point set in sequence as the next target position that the robot needs to reach, and use the A* and TEB algorithms to plan the global and local paths from the current position to the target position respectively, as shown in Figure 9 ; the green line represents the planned full-coverage path, and the red line represents the global and local paths planned by the A* algorithm and the TEB algorithm.

[0064] Then, calculate the speed control amount according to the TEB algorithm to control the robot to reach the target points in sequence. Then continue to extract the next target point from the target point set and repeat the above steps until the robot has reached all the target points and completed the cleaning work.

[0065] Repeat the above steps to construct for the arrangements of Figure 10 (a), (b), (c), (d) as shown in Figure 10(e), (f), (g), (h) show a navigation map with complete boundary information.

[0066] Inspired by the ideal embodiments of the present invention described above, through the above description, relevant staff can make various changes and modifications completely within the scope not deviating from the technical idea of this invention. The technical scope of this invention is not limited to the content in the specification, and its technical scope must be determined according to the scope of the claims.

Claims

1. A SLAM mapping and autonomous navigation method for a brush-type photovoltaic cleaning robot, characterized in that: The steps include: Step 1: The cleaning robot obtains image data through the depth camera; Step 2: Generate a 3D dense point cloud using the ORBSLAM2 algorithm, and separate the cleaned plane point cloud using the RANSAS algorithm; Step 3: Based on the cleaned plane point cloud, the NBEM algorithm and the LALSM algorithm are used to separate the boundary point cloud for the two arrangement modes of components with or without inner holes in the rooftop distributed photovoltaic power station; Step 4: Using the boundary point cloud as input, the octree algorithm is used to construct a two-dimensional grid map; Step 5: Convert the two-dimensional raster map into a pixel format image and preprocess it, use the boustrophedon algorithm to plan a full coverage cleaning path, and discretize the cleaning path into several target points; Step 6: Based on several target points, use A* and TEB algorithms to plan the paths between the target points, and control the robot to traverse the target points in sequence to complete the cleaning work.

2. The SLAM mapping and autonomous navigation method of a brush-type photovoltaic cleaning robot according to claim 1, characterized in that: In step 2, the RANSAS algorithm is used to separate the clean plane point cloud, and the given parameters are: The minimum number of points for the fitting model is 3, the distance threshold from each point to the fitting plane is 0.1-0.3m, and the number of iterations is 3-10.

3. The SLAM mapping and autonomous navigation method of a brush-type photovoltaic cleaning robot according to claim 1, characterized in that: Step 3 specifically includes the following steps: Step 3.1, determine the type of arrangement of photovoltaic modules in the rooftop distributed photovoltaic power station; Step 3.2: Taking the cleaned plane point cloud as input, for the arrangement with inner holes, the NBEM algorithm is used to separate the boundary point cloud of the cleaned plane with both inner and outer boundary information; for the arrangement without inner holes, the LALSM algorithm is used to separate the boundary point cloud of the cleaned plane containing only outer boundary information.

4. A SLAM mapping and autonomous navigation method for a disc-brush photovoltaic cleaning robot according to claim 1 or 3, characterized in that: The specific NBEM algorithm in step 3 is: First, the cleaned plane point cloud is used as input, and the NBEM algorithm is used to fit a global plane P and calculate the normal l of the global plane. n ; Secondly, take each point p in the point cloud i The surrounding finite points are used as input, and the NBEM algorithm is used to fit a local plane. And calculate the normal of the local plane Use it as the normal of each point in the swept plane point cloud; Finally, for each point p in the point cloud containing n points i , calculate the normal of the point and the global plane normal l n The angle between the boundary point cloud set P and the boundary point cloud set P is given an angle threshold θ. b for: 。 5. A SLAM mapping and autonomous navigation method for a disc-brush photovoltaic cleaning robot according to claim 1 or 3, characterized in that: The specific LALSM algorithm in step 3 is: First, the cleaned plane point cloud is projected onto the XOY plane to obtain a two-dimensional plane point cloud P p ; Secondly, according to the two-dimensional plane point cloud P p Each point p i Create the minimum enclosing rectangle R: in and They are point p i The horizontal and vertical coordinate values ​​of ; (a1,b1), (a1,b2), (a2,b2), (a2,b1) are the four vertices of the minimum enclosing rectangle respectively; Then, given the resolution res, the rectangle R is divided into (res+1)*(res+1) small rectangles, and the boundary point cloud P b for: of the x (a2-a1) / res,del y (b2-b1) / res In the formula, del x 、del y are the step sizes about the x and y axes respectively, are the horizontal and vertical coordinates of the i-th point respectively; are the minimum and maximum values ​​of the vertical coordinates of the point cloud in the jth interval respectively; are the minimum and maximum values ​​of the horizontal coordinates of the point cloud in the jth interval; P b is a collection of boundary point clouds.

6. The SLAM mapping and autonomous navigation method of a brush-type photovoltaic cleaning robot according to claim 1, characterized in that: Step 4 is as follows: using the boundary point cloud as input, using the octree algorithm to project each point in the boundary point cloud onto the XOY plane, and obtaining a two-dimensional grid map with a resolution of 0.01-0.05m.

7. The SLAM mapping and autonomous navigation method of a brush-type photovoltaic cleaning robot according to claim 1, characterized in that: Step 5 is as follows: First, the two-dimensional grid map is converted into a pixel format image, and the pixel format map is binarized; Secondly, the binary map is eroded and expanded; Then, the boustrophedon algorithm is used to plan a full coverage cleaning path on the eroded and expanded map; Finally, the cleaning path is discretized into a finite number of target points, and the connecting path between the target point sets is used as the theoretical full coverage cleaning path.

8. The SLAM mapping and autonomous navigation method of a brush-type photovoltaic cleaning robot according to claim 1, characterized in that: Step 6 is as follows: extract one point from the target point set in turn as the next target position that the robot needs to reach, and use the A* and TEB algorithms to plan the global and local paths from the current position to the target position respectively, and calculate the speed control amount according to the TEB algorithm to control the robot to reach the target points in turn until the robot has reached all the target points and completed the cleaning work.

Images