A method, system and device for completing negative obstacles in blind areas based on occupation grid

Abstract

The present invention discloses a method, system and device for completing negative obstacles in a blind area based on an occupied grid, which belongs to the field of environmental perception technology and is used to solve the current technical problem of high cost of negative obstacle perception. The method comprises the following steps: 1) obtaining a point cloud of the robot's surrounding environment and rasterizing the point cloud; 2) extracting ground information and negative obstacle information from the rasterized point cloud to obtain the boundary grid position of the negative obstacle and the corresponding ground point cloud; 3) performing hit information statistics from the negative obstacle boundary point cloud, and selecting the corresponding target grid update strategy according to the statistical results; 4) updating the occupancy probability according to the selected update strategy; 5) outputting negative obstacle information for the grid containing negative obstacle information in the laser radar blind area, and outputting complete obstacle information in combination with positive obstacle information. The present invention has the advantages of low cost and comprehensive obstacle information.

Description

Technical Field

[0001] The present invention mainly relates to the field of environmental perception technology, and specifically to a method, system and device for completing negative obstacles in a blind area based on an occupation grid. Background Art

[0002] In the automatic navigation of robots, environmental perception technology is a very critical technology. It directly tells the robot which places around it can travel and which places it cannot travel. Obstacle detection technology is an important component of environmental perception technology. In order to output obstacle information, robots are usually equipped with a laser radar on their top to enable the robot to see as far as possible. Then this configuration will cause the robot to have a certain blind spot in the near distance, so corresponding auxiliary perception equipment is usually installed in the robot's blind spot to extract obstacles. However, negative obstacles are usually not directly perceived because they are sunken under the ground. At the same time, installing auxiliary sensors also inadvertently increases the cost of the robot, so other technical means are needed to deal with negative obstacles in the blind spot.

[0003] The invention patent application with publication number CN109827585A and invention name "A method for quickly filling concave areas in grid maps" proposes a method for filling concave areas in grid maps. The implementation method is: first extract the occupied grid map, then select a unit from the map that is not an obstacle and has supporting force, and then expand horizontally from the left and right until there are obstacles on both the left and right sides. During the expansion process, the invention needs to continuously check whether there are obstacles in the upper and lower directions of the target grid, so that the expanded obstacles can correspond to the actual size. The invention is essentially a post-processing technology, which makes the processed negative obstacle size consistent with the actual size, and can solve the problem of broken, scattered and discontinuous negative obstacles, but it is only suitable for use in scenarios where the negative obstacles are within the field of view, and is not suitable for use in blind spots.

[0004] The invention patent application with the publication number CN113077551A and the invention name "Occupancy grid map construction method, device, electronic device and storage medium" proposes a method for generating an occupation map. The specific steps are: extracting obstacle point clouds and ground point clouds from point cloud information, fitting the driving road surface from the ground point cloud information and the ground physical size characteristics, and rasterizing; finally updating the occupation probability of obstacles on the rasterized road surface. The invention uses the occupancy map to update the occupancy probability of the positive obstacle category, and the ground point cloud is only used to segment the passable area in the invention. Although the idea of ​​occupancy grid is used, the operation of the occupancy probability is based on the number of points in the point cloud. Therefore, when the point cloud of the obstacle is considerable (this situation is very common in practice), the invention has already increased the occupancy probability to near the limit in one observation, and the subsequent occupancy probability update does not bring practical benefits; in addition, when there is a false alarm positive obstacle point cloud (for example, a positive obstacle output caused by a false alarm point cloud in a rainstorm), the method of the invention is unable to eliminate it. Therefore, the calculation method of the invention still has disadvantages. Summary of the invention

[0005] The technical problem to be solved by the present invention is: in view of the technical problems existing in the prior art, the present invention provides a method, system and device for completing negative obstacles in blind areas based on an occupied grid, which has low cost and ensures comprehensive obstacle information.

[0006] In order to solve the above technical problems, the technical solution proposed by the present invention is:

[0007] A method for completing negative obstacles in a blind area based on an occupied grid comprises the following steps:

[0008] 1) Obtain the point cloud of the robot's surrounding environment and perform rasterization on the point cloud;

[0009] 2) Extract ground information and negative obstacle information from the rasterized point cloud to obtain the boundary grid position of the negative obstacle and the corresponding ground point cloud;

[0010] 3) Perform hit information statistics from the negative obstacle boundary point cloud, and select the corresponding target grid update strategy based on the statistical results;

[0011] 4) Update the occupancy probability according to the selected update strategy;

[0012] 5) Output the negative obstacle information for the grids containing negative obstacle information in the blind area of ​​the laser radar, and output the complete obstacle information in combination with the positive obstacle information.

[0013] Preferably, the specific process of step 2) is:

[0014] The ground model algorithm is used to calculate the ground grid area from the rasterized point cloud to obtain the ground information;

[0015] The negative obstacle detection algorithm is used to calculate the boundary position of the negative obstacle from the rasterized point cloud to obtain the negative obstacle information;

[0016] The ground information and negative obstacle information are combined to extract the corresponding ground point cloud from the negative obstacle boundary grid.

[0017] Preferably, the specific process of using the ground model algorithm to calculate the ground grid area from the rasterized point cloud to obtain the ground information is as follows:

[0018] Convert the rasterized point cloud to the polar coordinate system; select a part of the area in front of the robot, and the height of all point clouds in the selected area meets the following requirements: the absolute value of the height is not higher than the preset value;

[0019] The height average of all point clouds in the selected area is used as the heuristic ground height empirical value. The ground height of the point cloud in each angle direction is estimated by taking the angle as the processing order: the ground height empirical estimate value plus the relaxation amount is used to obtain the ground height threshold of the target sector corresponding to the angle, and the point cloud whose height in the target sector does not exceed the ground height threshold is extracted as the valid ground point cloud, and the valid ground point cloud in the angle direction is obtained;

[0020] Polynomial curve fitting is performed in the angle direction, and the ground information of the area is obtained by obtaining the effective ground point cloud in each angle direction, wherein the ground information is the ground height.

[0021] Preferably, the specific process of using the negative obstacle detection algorithm to calculate the boundary position of the negative obstacle from the rasterized point cloud to obtain the negative obstacle information is:

[0022] Step by step, traverse the point cloud of each scan line, project the point cloud to the grid, and obtain the rasterized features of the scan line;

[0023] The grid position is used for continuity identification, and the point cloud whose height in the Cartesian coordinate system is below the grid ground height is selected, where the grid ground height in the Cartesian coordinate system corresponds to the ground height threshold in the polar coordinate system;

[0024] When the position of the ground point cloud on the grid is continuous, it is a normal road area. If the position of the point cloud projected to the grid jumps, it is considered that there are holes in the two projection grids. The two projection positions are marked as the boundary area of ​​the negative obstacle, and the corresponding ground point cloud is saved at the same time, that is, the point cloud of the negative obstacle boundary area;

[0025] Project all scan lines onto the grid, find out the grids whose height is lower than the ground height threshold by a certain value, and mark these grids as detectable negative obstacle areas.

[0026] Preferably, the specific process of extracting the corresponding ground point cloud from the negative obstacle boundary grid by combining the ground information and the negative obstacle information is as follows:

[0027] When a negative obstacle grid is observed, the initial observation confidence is added to the position. If there is a negative obstacle at the same position in the subsequent radar frame, the observation confidence is increased. If no negative obstacle area is observed at the position subsequently, the observation confidence is reduced. When the observation confidence is greater than a certain value, the corresponding ground point cloud is extracted from the negative obstacle boundary grid.

[0028] Preferably, the specific process of step 3) is:

[0029] For the grid with negative obstacles, the number of times this grid is hit by the ground point cloud is counted, and then the first update strategy, the second update strategy and the third update strategy are selected according to the number of times, specifically:

[0030] When the number of hits of the grid's point cloud reaches the sparsity threshold or does not exist, the subtractive update strategy is used;

[0031] When the number of hits of the grid point cloud is a normal count, the ordinary additive update strategy is selected;

[0032] When the number of hits exceeds a preset value, an aggressive additive update strategy is selected; wherein the operation of the aggressive additive update strategy is: using the ordinary additive update strategy multiple times in succession.

[0033] Preferably, in step 4), when the target grid is in a subtractive update strategy, the occupancy probability of the target grid is updated to decrease by a preset value; when the target grid is in a common additive update strategy, the occupancy probability of the target grid is updated to increase by a preset value.

[0034] Preferably, the specific process of step 5) is:

[0035] Set the confidence threshold and check whether the occupancy probability of the grid in the robot radar blind spot is lower than the confidence threshold; if the occupancy probability of the grid is lower than the confidence threshold, keep the grid vacant, otherwise change the grid to negative obstacle attribute;

[0036] After checking all the grids in the radar blind area, the negative obstacles in the blind area can be obtained, and the complete obstacle information can be output after being matched with the positive obstacle information.

[0037] The present invention also discloses a negative obstacle completion system in a blind area based on an occupation grid, comprising:

[0038] Point cloud generation module, used to obtain the point cloud of the robot's surrounding environment and perform raster processing on the point cloud;

[0039] The ground extraction module is used to extract ground information and negative obstacle information from the rasterized point cloud to obtain the boundary grid position of the negative obstacle and the corresponding ground point cloud;

[0040] Negative obstacle detection module, used to collect hit information statistics from the negative obstacle boundary point cloud, and select the corresponding target grid update strategy according to the statistical results;

[0041] A negative obstacle state update module, used to update the occupancy probability according to the selected update strategy;

[0042] The blind area result output module is used to output the negative obstacle information of the grid containing negative obstacle information in the blind area of ​​the laser radar, and output the complete obstacle information in combination with the positive obstacle information.

[0043] The present invention further discloses a computer device, comprising a memory and a processor, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the steps of the method described above are executed.

[0044] Compared with the prior art, the advantages of the present invention are:

[0045] The present invention makes use of the negative obstacle information of the blind spot where the robot resides, and by supplementing the negative obstacle information in the blind spot, it can reduce the number of auxiliary sensors, reduce the application cost of the robot, and solve the pain point of the robot being expensive and impractical, which has great economic benefits; only relying on the main sensor laser radar, it can realize the completion of negative obstacle information in the blind spot to ensure the comprehensiveness of the obstacle information; fully utilize the idea of ​​multi-frame fusion, do not overly rely on the contribution of negative obstacles in a certain frame, and ensure the stability of the negative obstacle information. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 The present invention is a flowchart of an embodiment of the method.

[0047] Figure 2 A block diagram of a system according to an embodiment of the present invention.

[0048] Figure 3 It is a rasterized schematic diagram of the scanning result in front of the laser radar in the present invention.

[0049] Figure 4 The effective field of view of the ring-scanning radar configured for the robot in the prior art, where (a) is the effective field of view of the top radar; (b) is the effective field of view of the front radar and the top radar combined.

[0050] Figure 5Schematic diagram of negative obstacle area processing in the present invention; (a) is the initial observation map; (b) is the negative obstacle area map; (c) is the negative obstacle boundary position marked with a higher confidence after three observations; (d) is a schematic diagram of the negative obstacle area that is not in the robot's field of view; (e) is a negative obstacle area situation map in the occupancy map; (f) is the result of using the occupancy map to complete the negative obstacle area in the blind spot; the upper box is the physical position of the robot, and the surrounding points are the output negative obstacle area near the vehicle body. DETAILED DESCRIPTION

[0051] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments.

[0052] like Figure 1 As shown, the negative obstacle completion method in the blind area based on the occupied grid of the embodiment of the present invention includes the steps of:

[0053] 1) Obtaining a point cloud of the robot's surrounding environment and performing rasterization processing on the point cloud to obtain a rasterized point cloud;

[0054] 2) Extract ground information and negative obstacle information from the rasterized point cloud to obtain the boundary grid position of the negative obstacle and the corresponding ground point cloud;

[0055] 3) Perform hit information statistics from the negative obstacle boundary point cloud, and select the corresponding target grid update strategy based on the statistical results;

[0056] 4) Update the occupancy probability according to the selected update strategy;

[0057] 5) Output the negative obstacle information for the grids containing negative obstacle information in the blind area of ​​the laser radar, and output the complete obstacle information in combination with the positive obstacle information.

[0058] The present invention makes use of the negative obstacle information of the blind spot where the robot resides. By supplementing the negative obstacle information in the blind spot, the number of auxiliary sensors can be reduced, the application cost of the robot can be reduced, and the pain point of the robot being expensive and impractical can be solved, which has great economic benefits. Only relying on the main sensor laser radar, the negative obstacle information can be supplemented in the blind spot to ensure the comprehensiveness of the obstacle information.

[0059] In addition, the idea of ​​multi-frame fusion is fully utilized, and the contribution of negative obstacles in a certain frame is not overly relied on, so as to ensure the stability of negative obstacle information. Specifically, suppose a certain grid is determined as a negative obstacle area by the obstacle detection result of one frame, but the grid is not detected as a negative obstacle in the results of the next two frames (or other frames, just for example). Therefore, its occupation probability is only an initial value. The next few frames believe that the grid does not belong to a negative obstacle, and the initial probability will be updated by the non-occupancy probability, so that its occupation probability is eliminated to the barrier-free attribute; so at the time of output, the negative obstacle attribute has been offset by the observation of the next few frames and becomes a free space; similarly, suppose a certain grid is determined as a negative obstacle area by the obstacle detection result of one frame, and an initial occupation probability is obtained; and the grid is detected as a negative obstacle in the results of the next two frames (or other frames, just for example), so its occupation probability will continue to be improved: its occupation probability is continuously superimposed on the basis of the initial probability to improve the occupation probability; so at the time of output, its occupation probability is sufficient to determine it as a negative obstacle grid, and finally marked as a negative obstacle area. Therefore, the above-mentioned processing scheme using the occupation probability method naturally has the idea of ​​multi-frame fusion.

[0060] In a specific embodiment, the specific process of step 2) is: using a ground model algorithm to calculate the ground grid area from the rasterized point cloud to obtain a ground calculation result; using a negative obstacle detection algorithm to calculate the boundary position of the negative obstacle from the rasterized point cloud to obtain a negative obstacle calculation result; combining the ground calculation result and the negative obstacle calculation result, extracting the corresponding ground point cloud from the negative obstacle grid.

[0061] Specifically: the ground model algorithm is used to calculate the ground grid area from the rasterized point cloud, and the specific process of obtaining the ground information is as follows:

[0062] Convert the rasterized point cloud to the polar coordinate system; select a part of the area in front of the robot, and the height of all point clouds in the selected area meets the following requirements: the absolute value of the height is not higher than the preset value;

[0063] The height average of all point clouds in the selected area is used as the heuristic ground height empirical value. The ground height of the point cloud in each angle direction is estimated by taking the angle as the processing order: the ground height empirical estimate value plus the relaxation amount is used to obtain the ground height threshold of the target sector corresponding to the angle, and the point cloud whose height in the target sector does not exceed the ground height threshold is extracted as the valid ground point cloud, and the valid ground point cloud in the angle direction is obtained;

[0064] Polynomial curve fitting is performed in the angle direction, and the ground information of the area is obtained by obtaining the effective ground point cloud in each angle direction, wherein the ground information is the ground height.

[0065] The specific process of using the negative obstacle detection algorithm to calculate the boundary position of the negative obstacle from the rasterized point cloud to obtain the negative obstacle information is as follows:

[0066] Step by step, traverse the point cloud of each scan line, project the point cloud to the grid, and obtain the rasterized features of the scan line;

[0067] The grid position is used for continuity identification, and the point cloud whose height in the Cartesian coordinate system is below the grid ground height is selected, where the grid ground height in the Cartesian coordinate system corresponds to the ground height threshold in the polar coordinate system;

[0068] When the position of the ground point cloud on the grid is continuous, it is a normal road area. If the position of the point cloud projected to the grid jumps, it is considered that there are holes in the two projection grids. The two projection positions are marked as the boundary area of ​​the negative obstacle, and the corresponding ground point cloud is saved at the same time, that is, the point cloud of the negative obstacle boundary area;

[0069] Project all scan lines onto the grid, find out the grids whose height is lower than the ground height threshold by a certain value, and mark these grids as detectable negative obstacle areas.

[0070] Specifically, combining the ground information and negative obstacle information, the specific process of extracting the corresponding ground point cloud from the negative obstacle boundary grid is as follows:

[0071] When a negative obstacle grid is observed, the initial observation confidence is added to the position. If there is a negative obstacle at the same position in the subsequent radar frame, the observation confidence is increased. If no negative obstacle area is observed at the position subsequently, the observation confidence is reduced. When the observation confidence is greater than a certain value, the corresponding ground point cloud is extracted from the negative obstacle boundary grid.

[0072] In a specific embodiment, the specific process of step 3) is: for a grid with negative obstacles, the number of times the grid is hit by the ground point cloud is counted, and then the first update strategy, the second update strategy and the third update strategy are selected according to the number of times, specifically: when the number of hits of the grid's point cloud is a sparsity threshold or there is no hit, the subtractive update strategy is selected; when the number of hits of the grid's point cloud is a normal count, the ordinary additive update strategy is selected; when the number of hits exceeds a predetermined set value, the active additive update strategy is selected; wherein, the operation of the active additive update strategy is: continuous use of multiple ordinary additive update strategies.

[0073] In a specific embodiment, in step 4), when the target grid is a subtractive update strategy, the occupancy probability of the target grid is updated to decrease by a preset value; when the target grid is a normal additive update strategy, the occupancy probability of the target grid is updated to increase by a preset value; when the target grid is an active additive update strategy, the occupancy probability of the target grid is updated to increase by multiple preset values, such as two; of course, three, four or more preset values ​​can also be selected according to actual conditions.

[0074] In a specific embodiment, the specific process of step 5) is: setting a confidence threshold, and checking whether the occupancy probability of the grid in the robot radar blind spot is lower than the confidence threshold; if the occupancy probability of the grid is lower than the confidence threshold, the vacant state of the grid is retained, otherwise the grid is changed to a negative obstacle attribute; after performing the check on all grids in the radar blind spot, the negative obstacles in the blind spot can be obtained, and the complete obstacle information is output after being matched with the positive obstacle information.

[0075] like Figure 2 As shown, an embodiment of the present invention provides a negative obstacle completion system in a blind area based on an occupation grid, comprising:

[0076] Point cloud generation module, used to obtain the point cloud of the robot's surrounding environment and perform raster processing on the point cloud;

[0077] The ground extraction module is used to extract ground information and negative obstacle information from the rasterized point cloud to obtain the boundary grid position of the negative obstacle and the corresponding ground point cloud;

[0078] Negative obstacle detection module, used to collect hit information statistics from the negative obstacle boundary point cloud, and select the corresponding target grid update strategy according to the statistical results;

[0079] A negative obstacle state update module, used to update the occupancy probability according to the selected update strategy;

[0080] The blind area result output module is used to output the negative obstacle information of the grid containing negative obstacle information in the blind area of ​​the laser radar, and output the complete obstacle information in combination with the positive obstacle information.

[0081] The negative obstacle completion system in a blind area based on an occupied grid of the present invention corresponds to the negative obstacle completion method described above, and also has the advantages described in the above completion method.

[0082] The above invention is further described in detail below in conjunction with a specific embodiment:

[0083] 1. Rasterize the environment around the robot;

[0084] Configure the laser radar so that the negative obstacle in front is within the laser radar's measurement field of view; use the laser radar to scan the robot's surrounding environment and rasterize the input point cloud;

[0085] Figure 3 This example shows how a LiDAR scan in front of you is converted into a raster image, where different colors are used to distinguish the measured heights of the grid and the small box is the body area of ​​the robot.

[0086] Depend on Figure 3 It can be seen that under this configuration, the real-time perception range of the robot is in front of the body (the effect is to achieve no blind spot perception), while the sides and rear of the robot near the body are blind spots because there is no real-time perception equipment in this area; although the mainstream configuration method is used to install a circular scanning radar on the top of the robot, such as Figure 4 As shown in (a), even if the effective fields of view of the two are superimposed, there is still a blind spot near the robot, such as Figure 4 As shown in (b), in order to effectively solve the problem that the robot may fall into negative obstacle areas such as potholes and cliffs when moving forward, the robot can use the effective perception information of the front in the past moments to complete the negative obstacles near the vehicle body, so as to achieve the purpose of effective obstacle avoidance, while saving the cost of deploying real-time sensors and improving commercial performance.

[0087] 2. Extracting ground information and negative obstacle information from the rasterized point cloud to obtain the boundary grid position of the negative obstacle (i.e. extracting the grid with negative obstacle edge information in the grid map) and the corresponding ground point cloud;

[0088] The ground model algorithm is used to calculate the ground area information from the point cloud; the negative obstacle detection algorithm is used to calculate the boundary position of the negative obstacle from the point cloud; the corresponding ground point cloud is extracted from the negative obstacle grid by combining the ground calculation results and the negative obstacle calculation results;

[0089] The ground model extraction method is as follows: the rasterized map point cloud is converted to a polar coordinate system, wherein the angle resolution is 2 degrees and the radial distance resolution is 50 cm. A portion of the area is selected from the area in front of the robot, and the height of all point clouds in the selected area satisfies: the absolute value of the height is not higher than 30 cm; the mean height of the point cloud in the selected area is used as the heuristic ground height empirical value; after obtaining the height empirical value, the ground height of the point cloud in each angle direction is estimated in the processing order of angle: first, the ground empirical threshold is calculated, and the rule is to increase the relaxation amount of 10 cm for every 2 m of radial distance. The ground height threshold of the target sector can be obtained by adding the ground height empirical estimate value to the relaxation amount, and the point cloud in the sector whose point cloud height does not exceed the ground height threshold is extracted as the effective ground point cloud; all effective ground point clouds in the angle direction are obtained; finally, polynomial curve fitting is performed using RANSAC in the angle direction (the fitting coefficient is 3 times); the ground height of all areas can be obtained by using this operation method for each angle direction.

[0090] After obtaining the ground height value, the next step is to extract the negative obstacle area (negative obstacle extraction algorithm), using two features: (1) the empty area of ​​a single scan line is the negative obstacle candidate area, and the edge point cloud of the negative obstacle area is extracted for marking; (2) the point cloud whose height measured by multiple scan lines in the same sector is less than 30 cm lower than the ground height of the sector is the point cloud that falls into the negative obstacle area, and the position of this part of the point cloud is directly used to mark it. The specific steps are:

[0091] The point cloud of each scan line is traversed step by step, and the point cloud is projected onto the grid to obtain the rasterized features of the scan line, including: the point cloud density, point cloud height, and grid position information of the grid. Continuity determination based only on the positions of two adjacent point clouds will lead to large changes in the results, so the grid position is used for continuity identification. Select the point cloud whose height is below the grid ground height (the previous ground estimation is converted to the polar coordinate system to obtain the ground height of the forward field of view, so the result can be converted to the Cartesian coordinate system to obtain the grid ground height in the same way). When the position of the ground point cloud on the grid is continuous, it is a normal road area (the projection position of the previous point cloud and the projection position of the next point cloud are within 2 pixels, and the resolution of the grid is 10cm / pixel). If the position of the point cloud projected to the grid jumps greatly (the grid projection position of the previous point cloud and the next point cloud differs by more than 2 pixels), it is considered that there are holes in the two projection grids, and the two projection positions are marked as the boundary area of ​​the negative obstacle, and the corresponding ground point cloud (negative obstacle boundary area point cloud) is saved at the same time.

[0092] Project all scan lines onto the grid, find out the grids whose height is more than 30cm lower than the estimated height on the ground, and mark these grids as detectable negative obstacle areas. In order to eliminate false alarms caused by erroneous detection of individual grids, the process also introduces spatial filtering technology, and the filter window size is set to 3x1. The grids left after filtering are then introduced into the spatial size verification (the horizontal size is at least 2 pixels, the vertical size is at least 2 pixels, and at least 3 negative obstacle marks can be found in the grid using a 5*2 inspection window (the mark in the first step is also valid). The point cloud in the result grid of this processing step is used as a detectable negative obstacle point cloud area.

[0093] After the above two steps, the next question is how to superimpose these negative obstacle boundaries (results of the first step) and areas (results of the second step) in multiple frames to identify the real negative obstacles. The idea of ​​the probability occupancy map is introduced here, and the initial occupancy probability of all grids is 0.5 (i.e., uncertain state). When a negative obstacle grid is observed, an observation is added to the position (given the initial occupancy probability, the probability value is 0.56). If there is a negative obstacle at the same position in the radar frame described later, the observation confidence is increased (add an occupied observation to the previous value, and its probability value is 0.65). If the subsequent observation does not observe a negative obstacle area at this position, the observation confidence is reduced (the previous occupancy probability minus one observation, the probability value is 0.58). The processing object here is negative obstacles, which come from the statistical characteristics of the grid. Therefore, the amount of point cloud data contained in the grid is very important information. If there are more point clouds, it can be increased to a larger observation confidence rate (according to the convention, add 2 or 3 times, that is, the effect of one observation is equivalent to the effect of observing 2 or 3 times); if the number of point clouds in the grid is small, it can be added once. Because it is operated in the global coordinate system, accurate coordinate values ​​are required. This is why the negative obstacle calculation results in the previous two steps output point clouds instead of grid positions.

[0094] 3. Count the “hit” information of the ground point cloud from the negative obstacle boundary point cloud and determine the update strategy of the target grid;

[0095] For the grid with negative obstacles, the number of times this grid is hit by the ground point cloud is counted, and the first update strategy, the second update strategy, and the third update strategy are used to update the occupancy probability:

[0096] (1) When the number of hits of the point cloud of the grid reaches the sparsity threshold or does not exist, the subtractive update strategy is used (when the number of point clouds contained in the grid is less than 3, the detection is considered invalid, so one observation is eliminated, that is, subtractive);

[0097] (2) When the number of hits of the grid point cloud is normal (the number of grid point clouds is between 4 and 10), the normal additive update strategy is selected;

[0098] (3) When the number of hits exceeds a preset value (when the number of point clouds in the grid is greater than 10), the active additive update strategy is selected; wherein the operation of the active additive update strategy is: using the general additive update strategy twice in succession.

[0099] 4. Based on the judgment results of the grid, use the corresponding update strategy to update the occupancy status.

[0100] Specifically, when the target grid is a subtractive update strategy, the probability of the target grid is updated to be reduced by a preset value (such as a probability value of 0.58); when the target grid is an additive update strategy, the probability of the target grid is increased by a preset value (such as a probability value of 0.65).

[0101] Note: The values ​​are not equal because 2 or 3 observations are enough to determine the existence of the target, while elimination requires more observations to prove the absence of the target.

[0102] Figure 5 Two operation effects are given, Figure 5 The circle in (b) is the negative obstacle area, and the small white dot inside is the marked boundary of the negative obstacle area. The state of these values ​​at the first observation is shown in (a); after 3 observations, the red mark has a higher confidence negative obstacle boundary position as shown in (c); when the negative obstacle area is not in the robot's field of view, as shown in (d); the negative obstacle area in the occupancy map is shown in (e), and the box indicates the position with high confidence; (f) is the result of using the occupancy map to complete the negative obstacle area in the blind spot. The upper box is the physical position of the robot, and the gray dot is the output negative obstacle area near the vehicle body.

[0103] The main idea is to detect and confirm the target (negative obstacle area) when it is within the field of view, and output the confirmed target when it is out of the field of view.

[0104] 5. Set the negative obstacle confidence threshold, output the negative obstacle information for the grids containing negative obstacle information in the blind area of ​​the lidar, and complete the negative obstacle information in the blind area into the obstacle map;

[0105] Set the confidence threshold, and check whether the occupancy probability of the grid in the robot radar blind area is lower than the confidence threshold. If the occupancy probability of the grid is lower than the threshold, the grid is kept vacant, otherwise the grid is changed to a negative obstacle attribute. After performing this check on all grids in the blind area, the obstacle map in the blind area can be obtained, and the complete obstacle information can be output after being combined with the positive obstacle information.

[0106] The embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored, and the computer program executes the steps of the above method when it is run by a processor. The embodiment of the present invention further provides a computer device, including a memory and a processor, and the memory stores a computer program, and the computer program executes the steps of the above method when it is run by the processor. The present invention implements all or part of the process in the above embodiment method, and can also be completed by instructing related hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above method embodiments when it is executed by the processor. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium, etc. The memory can be used to store computer programs and / or modules. The processor implements various functions by running or executing the computer programs and / or modules stored in the memory, and calling the data stored in the memory. The memory can include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other volatile solid-state storage devices, etc.

[0107] The above are only preferred embodiments of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technicians in this technical field, some improvements and modifications without departing from the principle of the present invention should be regarded as the protection scope of the present invention.

Claims

1. A method for completing negative obstacles in blind areas based on an occupied grid, characterized in that: Includes steps: 1) Obtain the point cloud of the robot's surrounding environment and rasterize the point cloud; 2) Extract ground information and negative obstacle information from the rasterized point cloud to obtain the boundary grid position of the negative obstacle and the corresponding ground point cloud; 3) Perform hit information statistics from the negative obstacle boundary point cloud, and select the corresponding target grid update strategy based on the statistical results; 4) Update the occupancy probability according to the selected update strategy; 5) Output the negative obstacle information of the grid containing negative obstacle information in the blind area of ​​the laser radar, and output the complete obstacle information in combination with the positive obstacle information; Combining ground information and negative obstacle information, the specific process of extracting the corresponding ground point cloud from the negative obstacle boundary grid is as follows: When a negative obstacle grid is observed, the initial observation confidence is added to the position. If there is a negative obstacle at the same position in the subsequent radar frame, the observation confidence is increased. If no negative obstacle area is observed at the position subsequently, the observation confidence is reduced. When the observation confidence is greater than a certain value, the corresponding ground point cloud is extracted from the negative obstacle boundary grid.

2. The method for completing negative obstacles in a blind area based on an occupation grid according to claim 1, characterized in that: The specific process of step 2) is as follows: The ground model algorithm is used to calculate the ground grid area from the rasterized point cloud to obtain the ground information; The negative obstacle detection algorithm is used to calculate the boundary position of the negative obstacle from the rasterized point cloud to obtain the negative obstacle information; The ground information and negative obstacle information are combined to extract the corresponding ground point cloud from the negative obstacle boundary grid.

3. The method for completing negative obstacles in a blind area based on an occupation grid according to claim 2, characterized in that: The specific process of calculating the ground grid area from the rasterized point cloud using the ground model algorithm to obtain the ground information is as follows: Convert the rasterized point cloud to the polar coordinate system; select a part of the area in front of the robot, and the height of all point clouds in the selected area meets the following requirements: the absolute value of the height is not higher than the preset value; The height average of all point clouds in the selected area is used as the heuristic ground height empirical value. The ground height of the point cloud in each angle direction is estimated by taking the angle as the processing order: the ground height empirical estimate value plus the relaxation amount is used to obtain the ground height threshold of the target sector corresponding to the angle, and the point cloud whose height in the target sector does not exceed the ground height threshold is extracted as the valid ground point cloud, and the valid ground point cloud in the angle direction is obtained; Polynomial curve fitting is performed in the angle direction, and the ground information of the area is obtained by obtaining the effective ground point cloud in each angle direction, wherein the ground information is the ground height.

4. The method for completing negative obstacles in a blind area based on an occupation grid according to claim 3, characterized in that: The specific process of using the negative obstacle detection algorithm to calculate the boundary position of the negative obstacle from the rasterized point cloud to obtain the negative obstacle information is as follows: Step by step, traverse the point cloud of each scan line, project the point cloud to the grid, and obtain the rasterized features of the scan line; The grid position is used for continuity identification, and the point cloud whose height in the Cartesian coordinate system is below the grid ground height is selected, where the grid ground height in the Cartesian coordinate system corresponds to the ground height threshold in the polar coordinate system; When the position of the ground point cloud on the grid is continuous, it is a normal road area. If the position of the point cloud projected to the grid jumps, it is considered that there are holes in the two projection grids. The two projection positions are marked as the boundary area of ​​the negative obstacle, and the corresponding ground point cloud is saved at the same time, that is, the point cloud of the negative obstacle boundary area; Project all scan lines onto the grid, find out the grids whose height is lower than the ground height threshold by a certain value, and mark these grids as detectable negative obstacle areas.

5. The method for completing negative obstacles in a blind area based on an occupation grid according to any one of claims 2 to 4, characterized in that: The specific process of step 3) is as follows: For the grid with negative obstacles, the number of times this grid is hit by the ground point cloud is counted, and then the first update strategy, the second update strategy and the third update strategy are selected according to the number of times, specifically: When the number of hits of the grid's point cloud reaches the sparsity threshold or does not exist, the subtractive update strategy is used; When the number of hits of the grid point cloud is a normal count, the ordinary additive update strategy is selected; When the number of hits exceeds a preset value, an aggressive additive update strategy is selected; wherein the operation of the aggressive additive update strategy is: using the ordinary additive update strategy multiple times in succession.

6. The method for completing negative obstacles in a blind area based on an occupation grid according to claim 5, characterized in that: In step 4), when the target grid is in a subtractive update strategy, the occupancy probability of the target grid is updated to decrease by a preset value; when the target grid is in a common additive update strategy, the occupancy probability of the target grid is updated to increase by a preset value.

7. The method for completing negative obstacles in a blind area based on an occupied grid according to claim 6, characterized in that: The specific process of step 5) is as follows: Set the confidence threshold and check whether the occupancy probability of the grid in the robot radar blind spot is lower than the confidence threshold; if the occupancy probability of the grid is lower than the confidence threshold, keep the grid vacant, otherwise change the grid to negative obstacle attribute; After checking all the grids in the radar blind area, the negative obstacles in the blind area can be obtained, and the complete obstacle information can be output after being matched with the positive obstacle information.

8. A system for completing negative obstacles in a blind area based on an occupation grid, used to execute the steps of the method for completing negative obstacles in a blind area based on an occupation grid as claimed in any one of claims 1 to 7, characterized in that: include: Point cloud generation module, used to obtain the point cloud of the robot's surrounding environment and perform raster processing on the point cloud; The ground extraction module is used to extract ground information and negative obstacle information from the rasterized point cloud to obtain the boundary grid position of the negative obstacle and the corresponding ground point cloud; Negative obstacle detection module, used to collect hit information statistics from the negative obstacle boundary point cloud, and select the corresponding target grid update strategy according to the statistical results; A negative obstacle state update module, used to update the occupancy probability according to the selected update strategy; The blind area result output module is used to output the negative obstacle information of the grid containing negative obstacle information in the blind area of ​​the laser radar, and output the complete obstacle information in combination with the positive obstacle information.

9. A computer device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: The computer program executes the steps of the method according to any one of claims 1 to 7 when executed by a processor.

Images