Method for effectively reducing number of freespace points of laser

Abstract

A method for effectively reducing the number of laser freespace points comprises the steps that firstly, rasterization processing is carried out on a laser point cloud to obtain grid indexes where all freespace points relative to the center of an own vehicle are located, then conversion is carried out through own vehicle posture information, all freespace points are transferred to fixed grids for internal maintenance, and finally, the number of the freespace points is reduced. The method comprises the following steps of: firstly, carrying out corresponding probability superposition on points in a fixed grid, then, carrying out outline drawing on the grid, specifically, traversing all grids with freespace points, judging whether the freespace grids are feature edge points or not by accessing the grid occupancy around the grids, and finally, sending laser freespace outline points. According to the method, the laser point cloud is rasterized, and the contour of the grid is depicted, so that the processing efficiency of the general freespace is greatly improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of autonomous driving, and in particular relates to a method for effectively reducing the number of laser freespace points. Background Art

[0002] In the autonomous driving system, environmental perception is an important function. It can obtain information about the surrounding environment through various sensors to facilitate path planning and control. We need to process the laser freespace point cloud data, that is, those points that are identified as spatial obstacles. These points can represent the drivable area, which is of great significance for path planning and control. However, in the current environmental perception analysis, we found a significant problem: the number of freespace points available for laser detection output is abnormally large. This situation may be caused by the high sensitivity of the sensor or the data processing algorithm. This excessive number of freespace points may have an impact on our overall autonomous driving performance, because the entire perception system needs to process and publish an excessive number of points, which leads to a decrease in the efficiency of the system in subsequent modules, and may even miss some potential obstacles or key areas due to time-consuming, which poses a certain risk. Therefore, in order to ensure the real-time results of the entire system, it is necessary to propose a new method for extracting freespace features based on contour strokes to reduce the amount of data and improve processing efficiency. The method first rasterizes the freespace points and divides the space into a grid-like structure for further analysis. Next, the probability accumulation technique is used to determine the key freespace points on the grid map. Finally, by checking the adjacent grids around each grid, it is determined whether the grid belongs to the outer contour of the freespace. This step is the key to identifying and defining the spatial boundary. The whole process not only improves the accuracy of the freespace feature, but also greatly reduces the amount of freespace that needs to be output. Summary of the invention

[0003] The technical problem to be solved by the present invention is to overcome the shortcomings of the prior art and provide a method for effectively reducing the number of laser freespace points, by traversing all grids with freespace points and accessing the grid occupancy around the grid to determine whether the freespace grid is a feature edge point, and finally sending the contour point of the laser freespace.

[0004] The present invention provides a method for effectively reducing the number of laser freespace points, comprising the following steps: Step S1. Rasterize the laser point cloud to obtain the grid index of each freespace point relative to the center of the ego vehicle at each moment, then update the internal grid pose according to the ego vehicle information, multiply the coordinates of the provided laser radar point in the ego vehicle system by the inverse of the resolution to obtain the coordinate result in the ego vehicle system, and finally fix the freespace point to the corresponding internal grid by taking the remainder; Step S2. Perform corresponding superposition probabilities on the points in the fixed grid; each frame of the laser point cloud will be rasterized. If the laser point cloud is transferred to a grid, the probability attribute of this grid will be increased by one. If there is no laser point cloud in the current grid, the probability will be reduced by 0.5. Finally, the probability greater than 2 will be output.

[0005] Step S3. By traversing all grids with freespace points, confirming the index of the current freespace point, then determining the indexes around the current index, judging the occupancy of the grids of the surrounding indexes, and finally de-rasterizing all grids marked as feature contour points and sending them out; Step S4. After the freespace points are rasterized, they are converted into fixed grids, and then probability accumulation is used to determine the key freespace points on the grid map; the grid is outlined by traversing all grids with freespace points and accessing the grid occupancy around the grid to determine whether the freespace grid is a feature edge point, and finally the contour point of the laser freespace is sent.

[0006] As a further technical solution of the present invention, in step S3, if the surrounding grids are occupied, the current grid is not identified as a feature contour point, and if the surrounding grids are not occupied, the current grid is identified as a feature contour point and marked.

[0007] Furthermore, in step S4, probability accumulation is used to determine the key freespace points on the grid map; each frame of the laser point cloud is rasterized, and if the laser point cloud is transferred to a grid, the probability attribute of the grid is increased by one, and if there is no laser point cloud in the current grid, the probability is reduced by 0.5, and finally the probability greater than 2 is output.

[0008] Furthermore, in step S5, the occupancy of the grids around the grid is accessed to determine whether the freespace grid is a feature edge point; since each laser point cloud will be rasterized, if the laser point cloud is transferred to a grid, the attribute of this grid is assigned to occupied. If the current point is occupied, the surrounding four points up, down, left and right will be traversed to see if they are occupied. If one of them is not occupied, this point is considered to be a boundary point.

[0009] The advantages of the present invention are: 1. Rasterize the laser freespace points and fix each result to the same grid map, which significantly improves efficiency 2. According to the outline of the overall freespace, the corresponding edge features are retained without changing the original point cloud freespace, which reduces the number of freespace points sent and improves the sending efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 is a rasterization schematic diagram of the present invention; Figure 2 A schematic diagram of a rasterized schematic diagram area of ​​the present invention; Figure 3 It is a schematic diagram of the stroke algorithm of the present invention; Figure 4 It is a schematic diagram of the method flow of the present invention. DETAILED DESCRIPTION

[0011] See also Figure 1-Figure 4 This embodiment provides a method of the present invention for effectively reducing the number of laser freespace points. The specific steps are as follows: like Figure 1 As shown in the figure, the laser point cloud is first rasterized to obtain the grid index of each freespace point relative to the center of the ego vehicle at each moment, and then the internal grid pose is updated according to the ego vehicle information. The coordinates of the provided laser radar point in the ego vehicle system are multiplied by the inverse of the resolution to obtain the coordinate result in the ego vehicle system. Finally, the fs point is fixed to the corresponding internal grid by taking the remainder.

[0012] Figure 2 The position of the grid where the freespace point is located after taking the remainder is given. That is, the position when the self-driving car starts at the initialization, and then the corresponding superposition probability is performed on the points in the fixed grid. Each frame of the laser point cloud will be rasterized. If the laser point cloud is transferred to a grid, the probability attribute of this grid will be increased by 1. If there is no laser point cloud in the current grid, the probability will be reduced by 0.5, and finally the probability greater than 2 will be output.

[0013] Next Steps Figure 3It is to outline the grid. Specifically, it traverses all the grids with freespace points. First, it confirms the index of the current freespace point, then determines the indexes around the current index, and judges the occupancy of the grids around the indexes. If the surrounding grids are occupied, the current grid is not identified as a feature contour point. If the surrounding grids are not occupied, the current grid will be identified as a feature contour point and marked. Finally, when sending, all the grids marked as feature contour points will be de-rasterized and sent out. Figure 4 The freespace points are rasterized and converted into fixed grids, and then the probability accumulation technique is used to determine the key freespace points on the grid map. Next, the grid is outlined by traversing all grids with freespace points and accessing the grid occupancy around the grid to determine whether the freespace grid is a feature edge point, and finally the contour point of the laser freespace is sent. Since each laser point cloud is rasterized, if the laser point cloud is transferred to a grid, the attribute of this grid is assigned to occupied. If the current point is occupied, the surrounding four points of up, down, left and right will be traversed to see if they are occupied. If one of them is not occupied, this point is considered to be a boundary point.

[0014] As the vehicle moves, the visible range of the map becomes larger and larger, and the corresponding array of superimposed probabilities will also become larger and larger; it is impossible to maintain an infinitely large array internally for probability superposition, so the modulo % is used to maintain a fixed-size map, which can also be understood as: no matter where the vehicle goes, the map index (+oo) near the vehicle will be normalized to the preset space [0,2048). The meaning of remainder is that when our fs point exceeds the range of position u under the self-vehicle system, the remainder method will be used to fix it back to the original grid. For example, (2048, 0) because our grid map size is 0 to 2047, that is to say, 2048 exceeds the grid map size we set. In this case, by taking the remainder, the fs at (2048, 0) becomes (1, 0) in our internal grid mapping. Each frame of the laser point cloud will be rasterized. If the laser point cloud is transferred to a grid, the probability attribute of this grid will be increased by one. If there is no laser point cloud in the current grid, the probability will be reduced by 0.5. Finally, the probability greater than 2 will be output, and then contour detection will be performed on it to reduce the output of the number of laser point clouds to reduce time consumption. The surrounding four points of up, down, left and right will be traversed to see if they are occupied. If one of them is not occupied, it is considered to be a boundary point. Outputting only boundary points can effectively reduce the number of laser point clouds.

[0015] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above specific embodiments. The above specific embodiments and the description in the specification are only for further illustrating the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of the present invention to be protected is defined by the claims and their equivalents.

Claims

1. A method for effectively reducing the number of laser freespace points, characterized in that: The following steps are included: Step S1. Rasterize the laser point cloud to obtain the grid index of each freespace point relative to the center of the ego vehicle at each moment, then update the internal grid pose according to the ego vehicle information, multiply the coordinates of the provided laser radar point in the ego vehicle system by the inverse of the resolution to obtain the coordinate result in the ego vehicle system, and finally fix the freespace point to the corresponding internal grid by taking the remainder; Step S2. The corresponding superposition probability is performed on the points in the fixed grid. Each frame of the laser point cloud will be rasterized. If the laser point cloud is transferred to a grid, the probability attribute of this grid is increased by 1. If there is no laser point cloud in the current grid, the probability is reduced by 0.

5. Finally, the probability greater than 2 is output; Step S3. By traversing all grids with freespace points, confirming the index of the current freespace point, then determining the indexes around the current index, judging the occupancy of the grids of the surrounding indexes, and finally de-rasterizing all grids marked as feature contour points and sending them out; Step S4. After the freespace points are rasterized, they are converted into fixed grids, and then probability accumulation is used to determine the key freespace points on the grid map; the grid is outlined by traversing all grids with freespace points and accessing the grid occupancy around the grid to determine whether the freespace grid is a feature edge point, and finally the contour point of the laser freespace is sent.

2. A method for effectively reducing the number of laser freespace points according to claim 1, characterized in that: In step S3, if the surrounding grids are occupied, the current grid is not identified as a feature contour point, and if the surrounding grids are not occupied, the current grid is identified as a feature contour point and marked.

3. A method for effectively reducing the number of laser freespace points according to claim 1, characterized in that: In step S4, probability accumulation is used to determine the key freespace points on the grid map; each frame of the laser point cloud is rasterized. If the laser point cloud is transferred to a grid, the probability attribute of the grid is increased by 1. If there is no laser point cloud in the current grid, the probability is reduced by 0.5, and finally the probability greater than 2 is output.

4. A method for effectively reducing the number of laser freespace points according to claim 1, characterized in that: In step S5, the grid occupancy around the grid is accessed to determine whether the freespace grid is a feature edge point. Since each laser point cloud will be rasterized, if the laser point cloud is transferred to a grid, the attribute of the grid is assigned to occupied. If the current point is occupied, the surrounding four points up, down, left and right will be traversed to see if they are occupied. If one of them is not occupied, this point is considered to be a boundary point.

Images