Obstacle detection denoising method and apparatus, and electronic device and storage medium

Abstract

Disclosed in the present application are an obstacle detection denoising method and apparatus, and an electronic device and a storage medium. The method comprises: acquiring non-terrestrial point clouds collected by a vehicle; on the basis of a vehicle coordinate system of the vehicle and a grid resolution, constructing a two-dimensional grid corresponding to a detection range, wherein the two-dimensional grid includes a plurality of cells; mapping the non-terrestrial point clouds to the two-dimensional grid; for each cell in the two-dimensional grid that has a non-terrestrial point mapped thereto, adding an attribute label to the cell on the basis of the non-terrestrial point mapped to the cell, wherein the attribute labelled to the cell at least includes one of an obstacle attribute and a noise attribute; and on the basis of the labelled attribute of each cell in the two-dimensional grid, dividing the non-terrestrial point clouds into obstacle point clouds and noise point clouds.

Description

Obstacle detection denoising method and device, electronic equipment and storage medium

[0001] Cross-reference to related applications

[0002] The present disclosure claims priority from Chinese Patent Application No. 202410895074.X filed on July 04, 2024 in China, the content of which is incorporated herein by reference in its entirety. TECHNICAL FIELD

[0003] The present application belongs to the technical field of obstacle detection, and particularly relates to an obstacle detection denoising method and device, an electronic equipment and a storage medium. BACKGROUND

[0004] In the field of autonomous vehicles, harsh conditions of the driving environment, such as dust, rain, fog, snow, etc., pose a serious challenge to obstacle detection. The original perception data of a laser radar has poor anti-interference ability in different environmental scenes and weather, such as when there is dust, rain, fog, snow, etc. in the environment. Especially in mine scenes, the mine vehicle working road is a non-paved road, which also brings a lot of dust, resulting in a large number of noise points in the obtained perception data, and the anti-interference ability of different brands of laser radars is quite different. Therefore, the obstacle detection result is not accurate. SUMMARY

[0005] The embodiments of the present application provide an obstacle detection denoising method and device, an electronic equipment and a storage medium, which can accurately distinguish obstacle point clouds and noise point clouds in driving environments such as dust, rain, fog, snow, etc., thereby improving the accuracy of obstacle detection.

[0006] In a first aspect, the embodiments of the present application provide an obstacle detection denoising method, comprising:

[0007] Obtaining a non-ground point cloud collected by a vehicle;

[0008] Based on a vehicle coordinate system of the vehicle and a grid resolution, a two-dimensional grid corresponding to a detection range is constructed, and the two-dimensional grid comprises a plurality of cells;

[0009] Mapping the non-ground point cloud to the two-dimensional grid;

[0010] For each cell in the two-dimensional grid that has a non-ground point mapped therein, based on the non-ground point mapped in the cell, an attribute label is added to the cell, and the attribute label of the cell at least comprises one of an obstacle attribute and a noise attribute, wherein the obstacle attribute is used to indicate that the non-ground point mapped in the cell is an obstacle point, and the noise attribute is used to indicate that the non-ground point mapped in the cell is a noise point;

[0011] The non-ground point cloud is divided into an obstacle point cloud and a noise point cloud based on attributes of each cell in the two-dimensional grid.

[0012] In a second aspect, an obstacle detection denoising apparatus is provided, including:

[0013] An acquisition module is configured to acquire a non-ground point cloud collected by a vehicle.

[0014] A grid construction module is configured to construct a two-dimensional grid corresponding to a detection range based on a vehicle coordinate system of the vehicle and a grid resolution, the two-dimensional grid including a plurality of cells.

[0015] A mapping module is configured to map the non-ground point cloud to the two-dimensional grid.

[0016] An attribute marking module is configured to, for each cell in the two-dimensional grid having the non-ground point mapped thereto, add an attribute mark to the cell based on the non-ground point mapped to the cell, the attribute mark of the cell including at least one of an obstacle attribute and a noise attribute, wherein the obstacle attribute is used to indicate that the non-ground point mapped to the cell is an obstacle point, and the noise attribute is used to indicate that the non-ground point mapped to the cell is a noise point.

[0017] A point cloud division module is configured to divide the non-ground point cloud into an obstacle point cloud and a noise point cloud based on the attribute mark of each cell in the two-dimensional grid.

[0018] In a third aspect, an electronic device is provided, including a processor and a memory storing computer program instructions.

[0019] The processor, when executing the computer program instructions, implements the obstacle detection denoising method of the first aspect.

[0020] In a fourth aspect, a computer readable storage medium is provided, the computer readable storage medium storing computer program instructions, the computer program instructions, when executed by a processor, implementing the obstacle detection denoising method of the first aspect.

[0021] In a fifth aspect, a computer program product is provided, the instructions in the computer program product being executed by a processor of an electronic device, so that the electronic device executes the obstacle detection denoising method of the first aspect.

[0022] This application discloses an obstacle detection denoising method, apparatus, electronic device, and storage medium. The method involves acquiring non-ground point clouds collected from a vehicle; constructing a two-dimensional grid corresponding to the detection range based on the vehicle's coordinate system and grid resolution, with multiple cells within the grid; mapping the non-ground point clouds to the two-dimensional grid; and for each cell in the two-dimensional grid mapped to a non-ground point, adding attribute markers to the cell based on the mapped non-ground point. The attributes of the cell markers include at least one of obstacle attributes and noise attributes. Based on the attributes of the cell markers in the two-dimensional grid, the non-ground point clouds are divided into obstacle point clouds and noise point clouds. According to this application, the non-ground point clouds can be accurately divided into obstacle point clouds and noise point clouds, thereby enabling obstacle detection based on the obstacle point clouds. This reduces the impact of noise point clouds on the detection results and improves the accuracy of obstacle detection. Attached Figure Description

[0023] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 is a schematic diagram of the sensing system provided in an embodiment of this application;

[0025] Figure 2 is a schematic flowchart of the obstacle detection and noise reduction method provided in an embodiment of this application;

[0026] Figure 3 is a schematic diagram of the obstacle detection and noise reduction device provided in an embodiment of this application;

[0027] Figure 4 is a schematic diagram of the structure of the electronic device provided in an embodiment of this application. Detailed Implementation

[0028] The features and exemplary embodiments of various aspects of this application will be described in detail below. To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only intended to explain this application and not to limit it. For those skilled in the art, this application can be implemented without some of these specific details. The following description of the embodiments is merely to provide a better understanding of this application by illustrating examples.

[0029] It should be noted that, in the present document, relational terms such as first and second and the like can be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises", "comprising", or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus. An element proceeded by "comprises... a" does not, without more constraints, exclude the presence of additional identical elements in the process, method, article, or apparatus that comprises the element.

[0030] To improve the accuracy of obstacle detection of a vehicle in a harsh environment, an embodiment of the present application provides an obstacle detection denoising method, device, electronic equipment and storage medium. First, the obstacle detection denoising method provided by the embodiment of the present application is introduced.

[0031] The obstacle detection denoising method provided by the embodiment of the present application can be applied to the obstacle detection scene of an autonomous vehicle, meets the real perception requirements of an autonomous driving perception system to the world, and is mainly used for identifying and filtering the laser radar noise point usage strategy that interferes with the real world in harsh environmental conditions such as dust, rain, fog, and snow, and is especially suitable for the dust scene on the non-paved road where the autonomous driving mine truck is located. Guarantee the real-time, clean, and accurate perception of the real world obstacles such as other vehicles, pedestrians, roadblocks, and the like by the autonomous vehicle perception system, and improve the robustness and reliability of the perception system.

[0032] The obstacle detection denoising method provided by the embodiment of the present application can be executed by a perception system, which can be installed on a vehicle. The perception system is composed of a hardware part and a software part, as shown in FIG. 1, wherein the hardware part of the perception system 100 includes but is not limited to a laser radar 101, a camera 102, a millimeter wave radar 103, a vehicle-mounted computing unit 104, a navigation device 105, an autonomous driving controller 106, and the like hardware devices, and the software part includes the perception module, the decision planning module, and the control module and the like software modules that are usually contained in the vehicle-mounted autonomous driving software platform not shown in FIG. 1.

[0033] The core scheme of the obstacle detection denoising method provided by the embodiment of the present application is embodied in the dust, rain, fog, and snow detection and filtering link of the perception module in the software part, which is responsible for real-time detection and identification of noise points in sensor data, and is responsible for filtering out these noise points to ensure the cleanliness and accuracy of the perception data. A technical scheme capable of improving the accuracy of obstacle detection in a dust, rain, fog, and snow environment is provided.

[0034] Referring to FIG. 2, a flowchart of an obstacle detection and denoising method provided by an embodiment of the present application is shown. As shown in FIG. 1, the method can include the following steps S21-S25, which are described in detail as follows.

[0035] S21. Obtain non-ground point cloud collected by a vehicle.

[0036] In some embodiments, the point cloud information is collected by a single or multiple lidars installed in the vehicle, and after ground segmentation of the point cloud information is performed by using a point cloud ground segmentation algorithm, ground point cloud and non-ground point cloud are obtained, and the obtained non-ground point cloud is obtained. The point cloud ground segmentation algorithm includes but is not limited to the DipG-Seg algorithm. The DipG-Seg algorithm uses a pixel-based image method, projects the point cloud onto two image planes, and performs projection image generation, image pre-segmentation, image fine segmentation, label voting, and other steps to complete the segmentation of the ground.

[0037] S22. Based on the vehicle coordinate system of the vehicle and the grid resolution, a two-dimensional grid corresponding to the detection range is constructed, and the two-dimensional grid includes a plurality of cells.

[0038] In some embodiments, after obtaining the non-ground point cloud, a two-dimensional grid can be constructed in a two-dimensional plane based on the vehicle coordinate system, and the two-dimensional grid is composed of a plurality of cells. The resolution cellSize of the two-dimensional grid, i.e., the size of each cell, can be set according to actual conditions. For example, the resolution of the two-dimensional grid can be set to 0.5 meters. The number of horizontal cells and the number of vertical cells of the two-dimensional grid can be determined by the detection range. The detection range can be the range that needs to be detected in real time for dust, rain, snow, and other environments. The size and position of the detection range can also be set according to actual conditions. For example, the detection range can be determined in the vehicle coordinate system with the maximum horizontal coordinate value rxMax being 75 meters, the minimum horizontal coordinate value rxMin being -20 meters, the maximum vertical coordinate value ryMax being 20 meters, and the minimum vertical coordinate value ryMin being -20 meters. The range determined in the vehicle coordinate system is the detection range, i.e., the range composed of the following four points (75, 20), (75, -20), (-20, 20), and (-20, 20) in the vehicle coordinate system is the detection range.

[0039] In the two-dimensional grid, a cell corresponds to a two-dimensional coordinate range in the vehicle coordinate system, and different cells correspond to different two-dimensional coordinate ranges.

[0040] S23. Map the non-ground point cloud to the two-dimensional grid.

[0041] In some embodiments, mapping the non-ground point cloud to the two-dimensional grid can include converting a three-dimensional coordinate of each point in the non-ground point cloud in a three-dimensional coordinate system to a two-dimensional coordinate in a vehicle coordinate system. For each point in the non-ground point cloud, comparing the two-dimensional coordinate of the point with a two-dimensional coordinate range corresponding to each cell in the two-dimensional grid, determining a two-dimensional coordinate range to which the two-dimensional coordinate of the point belongs, taking the cell corresponding to the two-dimensional coordinate range as the cell to which the point is mapped, and establishing an index relationship between the point and the cell to which the point is mapped, thereby mapping the point to the two-dimensional grid. In this way, the non-ground point cloud can be mapped to the two-dimensional grid.

[0042] Because the non-ground point cloud is an irregular discrete point, it is difficult to process and analyze. By mapping the non-ground point cloud to the two-dimensional grid, the non-ground point cloud data can be converted to a regular structure, facilitating further processing and analysis.

[0043] S24. For each cell in the two-dimensional grid that has a non-ground point mapped therein, adding an attribute label to the cell based on the non-ground point mapped into the cell, the attribute of the cell label including at least one of an obstacle attribute and a noise attribute, wherein the obstacle attribute is used to indicate that the non-ground point cloud mapped into the cell is an obstacle point cloud, and the noise attribute is used to indicate that the non-ground point cloud mapped into the cell is not an obstacle point cloud.

[0044] In some embodiments, each point in the non-ground point cloud has corresponding attribute information, including but not limited to intensity, height from the ground, etc., wherein the intensity represents the laser reflectivity of the object reflecting the point, and the height from the ground represents the distance between the object reflecting the point and the ground. Based on this, after mapping the non-ground point cloud to the two-dimensional grid, for each cell in the two-dimensional grid that has a non-ground point mapped therein, the attribute information corresponding to the cell can be determined based on the attribute information of the non-ground point mapped into the cell, and then the corresponding attribute label can be added to the cell based on the attribute information corresponding to the cell.

[0045] In some embodiments of the present application, the attribute information corresponding to the cell includes but is not limited to: minimum height of the cell from the ground, maximum height of the cell from the ground, average intensity of the cell, maximum intensity of the cell, number of points of the cell, number of low-intensity points of the cell, etc. Wherein, the minimum height of the cell from the ground is used to indicate the minimum height of the non-ground point cloud mapped into the cell from the ground; the maximum height of the cell from the ground is used to indicate the maximum height of the non-ground point cloud mapped into the cell from the ground; the average intensity of the cell is used to indicate the average intensity of the non-ground point cloud mapped into the cell; the maximum intensity of the cell is used to indicate the maximum intensity of the non-ground point cloud mapped into the cell; the number of points of the cell is used to indicate the number of non-ground point clouds mapped into the cell; and the number of low-intensity points of the cell is used to indicate the number of low-intensity points in the non-ground point cloud mapped into the cell.

[0046] In some embodiments of the present application, the attribute label of the cell can be added based on the non-ground points mapped into the cell by the following steps S241-S244:

[0047] S241. From the non-ground points mapped into the cell, low-intensity points are screened out.

[0048] In some embodiments, the low-intensity point refers to a point with an intensity lower than a certain intensity threshold, which can be set according to actual conditions.

[0049] In some embodiments of the present application, in the above step S241, the low-intensity points can be screened out from the non-ground point cloud mapped into the cell by the following steps S31-S32:

[0050] S31. For each non-ground point mapped into the cell, if the horizontal radial distance of the non-ground point does not belong to the set dynamic intensity radial threshold filtering range, the set minimum intensity threshold is taken as the intensity threshold corresponding to the non-ground point, and if the horizontal radial distance of the non-ground point belongs to the set dynamic intensity radial threshold filtering range, the intensity threshold corresponding to the non-ground point is determined based on the horizontal radial distance of the non-ground point, the maximum intensity threshold, the dynamic intensity radial threshold filtering range and the set minimum intensity threshold.

[0051] Because the intensity of the reflection of the non-ground point is different according to the distance between the object and the vehicle, and the intensity of the reflection is attenuated with the increase of the distance, if the same intensity threshold is set for the non-ground point, the low-intensity point determined finally is not accurate. Therefore, in some embodiments, the intensity threshold corresponding to the non-ground point is set based on the horizontal radial distance of the non-ground point, wherein the horizontal radial distance of the non-ground point is used to represent the distance between the object corresponding to the non-ground point and the vehicle. In this way, different intensity thresholds can be set for non-ground points with different horizontal radial distances, thereby improving the accuracy of the determined low-intensity point.

[0052] In some embodiments, the minimum intensity threshold intensity min and the maximum intensity threshold intensity max may be set in advance according to the actual situation, and the dynamic intensity radial threshold filtering range dy_intensity_filter_range_ is set in advance, wherein the dynamic intensity radial threshold filtering range is less than or equal to the detection range. For each non-ground point mapped into the cell, the horizontal radial distance of the non-ground point is calculated, wherein dis represents the horizontal radial distance of the point, x and y represent the coordinate values of the non-ground point in the vehicle coordinate system of the vehicle, when the horizontal radial distance of the non-ground point exceeds the dynamic intensity radial threshold filtering range, the set minimum intensity threshold intensity min is taken as the intensity threshold corresponding to the non-ground point, that is, intensity threshold = intensity min , wherein intensity threshold represents the intensity threshold corresponding to the non-ground point, when the horizontal radial distance of the non-ground point is within the dynamic intensity radial threshold filtering range, the intensity threshold corresponding to the non-ground point is calculated according to the function relationship between the intensity attenuation degree of the non-ground point corresponding to the object and the distance between the object and the vehicle, according to the following formula:

[0053] S32. In the case where the intensity of the non-ground point is less than the intensity threshold corresponding to the non-ground point, the non-ground point is determined as a low-intensity point.

[0054] In this way, the low-intensity point can be accurately screened out.

[0055] S242. The ratio of the number of low-intensity points to the number of non-ground points mapped into the cell is determined.

[0056] In some embodiments, the number of low-intensity points screened out in the above step S241 is determined, and the ratio of the number to the total number of non-ground points mapped into the cell is calculated.

[0057] S243. In a case where the ratio is greater than the proportion threshold, marking the cell as a noise attribute.

[0058] S244. In a case where the ratio is less than or equal to the proportion threshold, marking the cell as an obstacle attribute.

[0059] The proportion threshold can be set according to actual conditions.

[0060] Since the reflection intensity of the laser radar point cloud on dust, rain, fog and snow is small, when the proportion of low-intensity points in the non-ground points mapped into the cell is higher than a certain proportion threshold, it can be considered that the non-ground points mapped into the cell are reflected by dust, rain, fog and snow, and thus the non-ground points are marked as a noise attribute, otherwise, the non-ground points are marked as an obstacle attribute.

[0061] S25. Based on the attributes marked in each cell in the two-dimensional grid, the non-ground point cloud is divided into an obstacle point cloud and a noise point cloud.

[0062] In some embodiments, the obstacle point cloud is a point cloud used for obstacle detection, and the noise point cloud is a point cloud that needs to be removed when performing obstacle detection. Since real obstacles and noise have different characteristics, and different attributes marked in the cells can represent different characteristics. Therefore, after completing the attribute marking of the cells in the two-dimensional grid, the non-ground point cloud can be divided into an obstacle point cloud and a noise point cloud based on the attributes marked in the cells.

[0063] In some embodiments of the present application, the non-ground points mapped in the cells marked as obstacle attributes can be divided into an obstacle point cloud, and the non-ground points mapped in the cells marked as noise attributes can be divided into a noise point cloud.

[0064] The obstacle detection and noise removal method provided in the embodiments of the present application acquires a non-ground point cloud collected by a vehicle; constructs a two-dimensional grid corresponding to a detection range based on a vehicle coordinate system of the vehicle and a grid resolution, the two-dimensional grid comprising a plurality of cells; maps the non-ground point cloud to the two-dimensional grid; for each cell in the two-dimensional grid having non-ground points mapped therein, adds an attribute mark to the cell based on the non-ground points mapped into the cell, the attribute marked in the cell including at least one of an obstacle attribute and a noise attribute; and divides the non-ground point cloud into an obstacle point cloud and a noise point cloud based on the attributes marked in each cell in the two-dimensional grid. According to the embodiments of the present application, the non-ground point cloud can be divided into an obstacle point cloud and a noise point cloud, so that obstacle detection can be performed based on the obstacle point cloud, and thus the influence of the noise point cloud on the detection result can be reduced, and the obstacle detection accuracy can be improved.

[0065] In some embodiments, there can be a problem of low accuracy when only adding attribute labels to the cells according to the non-ground points mapped into the cells. In view of this, in order to improve the accuracy of the attribute labels and thus improve the accuracy of the point cloud division, before the step S25, the attribute labels of the cells can be adjusted based on a set smoothing strategy.

[0066] In some embodiments, in the step S24, when adding attribute labels to the cells based on the non-ground points mapped into the cells, the attribute labels of the cells can further include one of a high attribute and a low attribute, and one of a far distance attribute and a near distance attribute, wherein the high attribute is used to indicate that there is a non-ground point with a height greater than a first height threshold in the non-ground points mapped into the cell, the low attribute is used to indicate that the heights of the non-ground points mapped into the cell are all less than or equal to the first height threshold, the far distance attribute is used to indicate that the minimum height of the non-ground points mapped into the cell is greater than a second height threshold, and the near distance attribute is used to indicate that the minimum height of the non-ground points mapped into the cell is less than or equal to the second height threshold. The first height threshold and the second height threshold can be set according to actual conditions.

[0067] Correspondingly, in the step S24, the following steps can also be performed.

[0068] determining whether the heights of the non-ground points mapped into the cell are all less than the first height threshold;

[0069] in the case where the heights of the non-ground points mapped into the cell are all less than the first height threshold, labeling the cell as the low attribute;

[0070] in the case where there is a non-ground point with a height greater than or equal to the first height threshold in the non-ground points mapped into the cell, labeling the cell as the high attribute;

[0071] determining the minimum height of the non-ground points mapped into the cell from the ground;

[0072] in the case where the minimum height is less than the second height threshold, labeling the cell as the near distance attribute;

[0073] in the case where the minimum height is greater than or equal to the second height threshold, labeling the cell as the far distance attribute.

[0074] Based on this, the attribute labels of the cells can be adjusted through the following steps S41-S49.

[0075] S41. In the case where the cell is labeled as the noise attribute and the low attribute, traversing the first neighborhood cells located within a first neighborhood radius of the cell.

[0076] Here, the first neighborhood radius can be set according to actual conditions, for example, the first neighborhood radius can be set to 8 cells, so that a plurality of cells located in the 8-neighborhood of the cell periphery can be taken as the first neighborhood cells, thereby traversing the plurality of first neighborhood cells.

[0077] S42. In the first neighborhood cell, if the number of first neighborhood cells marked as the noise attribute is less than the first number threshold, update the noise attribute of the cell to the obstacle attribute.

[0078] Here, the first number threshold can be set according to actual conditions, for example, it can be 3.

[0079] In some embodiments, when traversing the plurality of first neighborhood cells, it can be determined whether each first neighborhood cell is marked as the noise attribute, thereby determining the first neighborhood cells marked as the noise attribute from the plurality of first neighborhood cells, and determining the number of first neighborhood cells marked as the noise attribute, and comparing the number with the first number threshold. If the number is less than the first number threshold, it can be considered that the cell should be marked as the obstacle attribute, thereby updating the noise attribute of the cell to the obstacle attribute.

[0080] S43. In the case where the cell is marked as the far distance attribute, and the number of non-ground point clouds mapped into the cell is greater than the second number threshold, traverse the second neighborhood cells located within the second neighborhood radius of the cell periphery.

[0081] Here, the second number threshold and the second neighborhood radius can be set according to actual conditions. For example, the second neighborhood radius can be set to 6 cells, so that a plurality of cells located in the 6-neighborhood of the cell periphery can be taken as the second neighborhood cells, thereby traversing the plurality of second neighborhood cells.

[0082] S44. In the second neighborhood cell, if the number of second neighborhood cells marked as the near distance attribute is greater than the third number threshold, update the far distance attribute of the cell to the near distance attribute.

[0083] Here, the third number threshold can be set according to actual conditions.

[0084] In some embodiments, when traversing the plurality of second neighborhood cells, it can be determined whether each second neighborhood cell is marked as the near distance attribute, thereby determining the second neighborhood cells marked as the near distance attribute from the plurality of second neighborhood cells, and determining the number of second neighborhood cells marked as the near distance attribute, and comparing the number with the third number threshold. If the number is greater than the third number threshold, it can be considered that the cell should be marked as the near distance attribute, thereby updating the far distance attribute of the cell to the near distance attribute.

[0085] S45. Traversing third neighborhood cells located within a third neighborhood radius outside the vehicle body outline.

[0086] In some embodiments of the present application, the vehicle body outline can be determined according to the index of the four corners of the vehicle in the two-dimensional grid, wherein the four corners of the vehicle include the left front corner point, the right front corner point, the left rear corner point and the right rear corner point of the vehicle. Wherein for each corner point, its index in the two-dimensional grid can be determined based on the following formula: indexX = floor((x-rxMin) / cellSize) indexY = floor((y-ryMin) / cellSize)

[0087] In the formula, floor represents the down rounding calculation, indexX and indexY represent the index coordinates of the corner point in the two-dimensional grid, and x, y represent the coordinates of the vehicle corner point in the vehicle coordinate system. Wherein, the index coordinates of the point in the two-dimensional grid are used to indicate the relative position of the cell to which the point is mapped in the two-dimensional grid, that is, to indicate the row and column of the cell in the two-dimensional grid.

[0088] Here, the third neighborhood radius can be set according to actual conditions, for example, the third neighborhood radius can be set to 6 cells, so that a plurality of cells located in the 6-neighborhood of the vehicle body outline outside the vehicle body outline can be taken as third neighborhood cells, thereby traversing a plurality of third neighborhood cells.

[0089] S46. In the case that there is a third neighborhood cell marked with an obstacle attribute in the third neighborhood cell, traversing fourth neighborhood cells located within a fourth neighborhood radius of the third neighborhood cell marked with an obstacle attribute.

[0090] In some embodiments, when traversing a plurality of third neighborhood cells, it can be determined whether each third neighborhood cell is marked with an obstacle attribute, thereby determining a third neighborhood cell marked with an obstacle attribute from the plurality of third neighborhood cells. In the case that a third neighborhood cell marked with an obstacle attribute is determined, traversing fourth neighborhood cells located within a fourth neighborhood radius of the third neighborhood cell marked with an obstacle attribute.

[0091] Here, the fourth neighborhood radius can be set according to actual conditions, for example, the fourth neighborhood radius can be set to 6 cells, so that a plurality of cells located in the 6-neighborhood of the third neighborhood cell marked with an obstacle attribute can be taken as fourth neighborhood cells, thereby traversing a plurality of fourth neighborhood cells.

[0092] S47. In the fourth neighborhood cell, if the number of fourth neighborhood cells marked as the noise attribute is greater than the fourth number threshold, update the obstacle attribute of the third neighborhood cell marked as the obstacle attribute in the third neighborhood cell to the noise attribute.

[0093] Here, the fourth number threshold can be set according to actual conditions.

[0094] In some embodiments, when traversing the plurality of fourth neighborhood cells, it can be judged whether each fourth neighborhood cell is marked as the noise attribute, so as to determine the fourth neighborhood cell marked as the noise attribute from the plurality of fourth neighborhood cells, and determine the number of fourth neighborhood cells marked as the noise attribute, compare the number with the fourth number threshold, if the number is greater than the fourth number threshold, it can be considered that the third neighborhood cell marked as the obstacle attribute should actually be the noise attribute, so as to update the long-distance attribute of the third neighborhood cell marked as the obstacle attribute to the short-distance attribute.

[0095] S48. Traverse the fifth neighborhood cell located within the fifth neighborhood radius outside the vehicle body contour.

[0096] Here, the fifth neighborhood radius can be set according to actual conditions, for example, the fifth neighborhood radius can be set to 6 cells, so that the plurality of cells located in the 6-neighborhood around the vehicle body contour outside can be set as the fifth neighborhood cell, so as to traverse the plurality of fifth neighborhood cells.

[0097] S49. In the fifth neighborhood cell, if there is a fifth neighborhood cell marked as the obstacle attribute, if the number of non-ground point clouds mapped into the fifth neighborhood cell marked as the obstacle attribute is less than the fifth number threshold, and the minimum height of the non-ground point clouds mapped into the fifth neighborhood cell marked as the obstacle attribute is greater than the third height threshold, update the obstacle attribute of the fifth neighborhood cell marked as the obstacle attribute to the noise attribute.

[0098] Here, the fifth number threshold and the third height threshold can be set according to actual conditions.

[0099] In some embodiments, when traversing the plurality of fifth neighborhood cells, it can be determined whether each fifth neighborhood cell is marked as the obstacle attribute, so as to determine the fifth neighborhood cell marked as the obstacle attribute from the plurality of fifth neighborhood cells, and determine the number of the fifth neighborhood cells marked as the obstacle attribute, compare the number with the fifth number threshold, if the number is less than the fifth number threshold, and the minimum height of the non-ground point cloud mapped to the fifth neighborhood cell marked as the obstacle attribute is greater than the third height threshold, it can be considered that the fifth neighborhood cell marked as the obstacle attribute should actually be the noise attribute, so as to update the long-distance attribute of the fifth neighborhood cell marked as the obstacle attribute to the short-distance attribute.

[0100] Through the above attribute adjustment, the accuracy of the attributes marked by each cell can be improved, thereby improving the accuracy of the point cloud division, and further improving the accuracy of the obstacle detection.

[0101] In some embodiments, in the above step S25, based on the attributes marked by each cell in the two-dimensional grid, the non-ground point cloud can be divided into the obstacle point cloud and the noise point cloud, which can include:

[0102] For each non-ground point in each non-ground point cloud, the following steps S51-S59 are respectively performed.

[0103] S51. Determine whether the non-ground point is mapped to the two-dimensional grid.

[0104] In some embodiments, the range of the point cloud collected by the laser radar is generally larger than the detection range, so there can be non-ground points that are not mapped to the two-dimensional grid.

[0105] In some embodiments of the present application, the non-ground point not mapped to the two-dimensional grid is a non-ground point whose corresponding two-dimensional coordinates do not belong to the two-dimensional coordinate range corresponding to any cell in the two-dimensional grid.

[0106] S52. In the case that the non-ground point is not mapped to the two-dimensional grid, divide the non-ground point into the obstacle point cloud.

[0107] In the case that the non-ground point is not mapped to the two-dimensional grid, it can be considered that the non-ground point is a point reflected back by an object outside the detection range, and compared with the objects within the detection range, the objects outside the detection range are farther away from the vehicle, and the reflection intensity of dust, rain, fog and snow is relatively small, so it is difficult for dust, rain, fog and snow located outside the detection range to have a point reflected back, and because the distance is too far, it does not affect the driving decision and planning of the current vehicle, so the non-ground point not mapped to the two-dimensional grid can be considered not to be reflected back by dust, rain, fog and snow. Therefore, in the case that the non-ground point is not mapped to the two-dimensional grid, it can be considered that the non-ground point is a point reflected back by a real obstacle, so as to divide it into the obstacle point cloud.

[0108] S53. In the case that the non-ground point is mapped into a cell labeled as an obstacle cell with an obstacle attribute, or a noise cell with a noise attribute, if the non-ground point is mapped into the obstacle cell, S54-S57 are performed, and if the non-ground point is mapped into the noise cell, S58-S59 are performed.

[0109] S54. If the obstacle cell is labeled as a long-distance attribute, and the average intensity of the non-ground points mapped into the obstacle cell is less than a first intensity threshold, the non-ground points are divided into a noise point cloud.

[0110] S55. If the minimum height of the obstacle cell from the ground is greater than a fourth height threshold, and the average intensity of the non-ground points mapped into the obstacle cell is less than the first intensity threshold, the non-ground points are divided into a noise point cloud, the minimum height of the cell from the ground being the minimum height of the non-ground points mapped into the cell from the ground.

[0111] S56. If the minimum height of the obstacle cell from the ground is greater than the fourth height threshold, and the number of non-ground points mapped into the obstacle cell is less than a sixth number threshold, the non-ground points are divided into a noise point cloud.

[0112] S57. If none of the conditions in S54-S56 are met, the non-ground points are divided into an obstacle point cloud.

[0113] S58. If the minimum height of the noise cell from the ground is less than a fifth height threshold, and the number of non-ground points mapped into the noise cell is greater than a seventh number threshold, the non-ground points are divided into an obstacle point cloud.

[0114] S59. If the condition in S58 is not met, the non-ground points are divided into a noise point cloud.

[0115] By dividing the point cloud in the above manner, the accuracy is higher than that of only relying on the attribute of the cell label.

[0116] In some embodiments, in order to improve the accuracy of point cloud division, the above step S24, adding an attribute label to the cell based on the non-ground point cloud mapped into the cell, can further include:

[0117] performing clustering processing on the non-ground point cloud to obtain a plurality of point cloud clusters;

[0118] performing straight line fitting on the point cloud in each point cloud cluster respectively to obtain a straight line segment set corresponding to each point cloud cluster;

[0119] In the corresponding set of line segments, the point cloud cluster containing the line with a length greater than the length threshold is taken as the point cloud cluster corresponding to the real obstacle;

[0120] If the point in the point cloud cluster corresponding to the real obstacle is mapped to the cell, the cell is marked as a regular attribute, and the regular attribute is used to indicate that the non-ground point mapped to the cell contains the point corresponding to the real obstacle.

[0121] Here, is used to extract some regular targets such as vehicle, wall, and edge features. First, the non-ground point cloud is clustered, and a plurality of point cloud clusters, i.e., point cloud clusters, are obtained after clustering. For each point cloud cluster, line fitting is performed to obtain a set of line segments of each point cloud cluster. When there is a line segment with a length greater than a length threshold in the set of line segments, the point cloud cluster is considered as a point cloud corresponding to a real obstacle, and a regular attribute is added to the cell corresponding to the point cloud cluster. The length threshold can be set according to actual conditions.

[0122] Correspondingly, in the step S25, based on the adjusted attribute label of the cell, the non-ground point cloud is divided into obstacle point cloud and noise point cloud, and further comprising:

[0123] Before determining whether the non-ground point is mapped to the obstacle cell marked as the obstacle attribute or the noise cell marked as the noise attribute, it is determined whether the cell to which the non-ground point is mapped is marked as the regular attribute;

[0124] In the case where the cell to which the non-ground point is mapped is marked as the regular attribute, if the number of non-ground point clouds contained in the cell to which the non-ground point is mapped is greater than an eighth number threshold, the point is marked as an obstacle point cloud;

[0125] In the case where the cell to which the non-ground point is mapped is not marked as the regular attribute, the step of determining whether the non-ground point is mapped to the obstacle cell marked as the obstacle attribute or the noise cell marked as the noise attribute is performed.

[0126] In some embodiments, the non-ground point cloud mapped in the cell marked as the regular attribute is the point cloud corresponding to the real obstacle, and therefore, the non-ground point mapped in the cell marked as the regular attribute can be directly divided into the obstacle point cloud, without further dividing the non-ground point mapped in the cell based on other attributes of the cell. In this way, the non-ground point cloud can be avoided to be incorrectly divided based on the obstacle attribute or the noise attribute, thereby improving the accuracy of point cloud division.

[0127] In some embodiments, in order to improve the accuracy of point cloud division, in the step S24, based on the non-ground point mapped to the cell, the attribute label of the cell is added, and further comprising:

[0128] obtain vehicle on-board unit information, the vehicle on-board unit information comprising relevant information of other vehicles, the other vehicles being vehicles capable of communicating with the vehicle on-board unit of the vehicle;

[0129] determine, based on the vehicle on-board unit information, a corresponding range of the other vehicles in a vehicle coordinate system of the vehicle;

[0130] determine whether the cell is located within the corresponding range of the other vehicles in the vehicle coordinate system of the vehicle;

[0131] in a case where the cell is located within the corresponding range of the other vehicles in the vehicle coordinate system of the vehicle, mark the cell as a vehicle attribute.

[0132] In some embodiments, the vehicle on-board unit OBU of the vehicle can obtain relevant information of other OBU-containing vehicles in the system in real time, and based on the pose of the vehicle, the corresponding range of each other vehicle in the vehicle coordinate system of the vehicle can be determined based on the vehicle coordinate system of the vehicle.

[0133] In some embodiments of the present application, when determining the corresponding range of the other vehicles in the vehicle coordinate system of the vehicle, the coordinates of the center point and the coordinates of the four corner points of the other vehicles in the UTM coordinate system, i.e., the Universal Transverse Mercator Grid System, can be converted to the vehicle coordinate system of the vehicle. Then, the range determined in the vehicle coordinate system of the vehicle based on the converted coordinates of the center point and the coordinates of the four corner points is taken as the range of the other vehicles in the vehicle coordinate system of the vehicle.

[0134] When converting the center point position of the other vehicle from the UTM coordinate system to the vehicle coordinate system of the vehicle, the position of the other vehicle at the current frame point cloud timestamp can be corrected first:

[0135] obu_utm_pos_cur=obu_utm_pos+obu_yaw*obu_vel*obu_dt

[0136] In the formula, obu_utm_pos_cur is the position of the other vehicle at the current point cloud timestamp, obu_utm_pos is the UTM coordinate of the center point of the other vehicle, obu_vel is the speed of the other vehicle, obu_yaw is the yaw angle of the other vehicle, and obu_dt is the time difference between the point cloud time and the time of the other vehicle.

[0137] After the above correction is completed, the center point coordinate of the other vehicle is converted from the UTM coordinate system to the vehicle coordinate system of the vehicle by the following formula:

[0138] obu_veh_pos=q_cur_inv*(obu_utm_pos_cur-veh_utm_pos)

[0139] obu_corner_utm_pos_cur = obu_utm_pos_cur + obu_yaw * obu_corner_offset

[0140] The positions of the four corner points of the other vehicle in the vehicle coordinate system of the host vehicle are also calculated using the same conversion method.

[0141] The current positions of the four corner points of the other vehicle in the UTM coordinate system are as follows:

[0142] obu_corner_utm_pos_cur = obu_utm_pos_cur + obu_yaw * obu_corner_offset

[0143] In the formula, obu_corner_utm_pos_cur represents the current positions of the four corner points of the other vehicle in the UTM coordinate system, and obu_corner_offset represents the offset of the four corner points of the other vehicle relative to the center point.

[0144] The four corner points of the other vehicle are converted to the vehicle coordinate system of the host vehicle by the following formula:

[0145] obu_corner_utm_pos_cur = obu_utm_pos_cur + obu_yaw * obu_corner_offset

[0146] In the formula, obu_corner_utm_pos_cur represents the current positions of the four corner points of the other vehicle in the UTM coordinate system, and obu_corner_offset represents the offset of the four corner points of the other vehicle relative to the center point.

[0147] In some embodiments of the present application, in order to quickly determine the cells located in the corresponding range of the other vehicle in the vehicle coordinate system, a screening range smaller than the detection range can be determined first, which contains the range of the other vehicle in the vehicle coordinate system of the host vehicle. The cells located outside the screening range are not located in the range of the other vehicle in the vehicle coordinate system of the host vehicle. Therefore, it is only necessary to determine whether the cells located in the screening range are located in the range of the other vehicle in the vehicle coordinate system of the host vehicle. In this way, the calculation amount can be reduced, and the cells located in the corresponding range of the other vehicle in the vehicle coordinate system can be quickly determined.

[0148] In determining the screening range, four extreme values can be determined according to the coordinates of the four corner points of the other vehicle in the vehicle coordinate system, the four extreme values being the maximum horizontal coordinate max x, the minimum horizontal coordinate min x, the maximum vertical coordinate max y and the minimum vertical coordinate min y in the coordinates of the four corner points in the vehicle coordinate system, and four extreme points can be determined based on the four extreme values, being (max x, max y), (max x, min y), (min x, max y) and (min x, min y).

[0149] Then the four corner points are mapped from the vehicle coordinate system to the two-dimensional grid by the following mapping formula to obtain the index coordinates p1(x1, y1), p2(x2, y2), p3(x3, y3) and p4(x4, y4) of the four corner points in the two-dimensional grid: indexX = floor((x - rxMin) / cellSize) indexY = floor((y - ryMin) / cellSize)

[0150] In calculating the index coordinates of the four corner points in the two-dimensional grid based on the above mapping formula, x and y in the above mapping formula represent the position of each corner point in the vehicle coordinate system.

[0151] Similarly, the index coordinates min_indexX, max_indexX, min_indexY and max_indexY of the four extreme points in the two-dimensional grid are calculated based on the above mapping formula. In calculating the index coordinates of the four extreme points in the two-dimensional grid based on the above mapping formula, x represents the minimum horizontal coordinate min x or the maximum horizontal coordinate max x, and y represents the minimum vertical coordinate min y or the maximum vertical coordinate max y.

[0152] The quadrilateral range composed of p1(x1, y1), p2(x2, y2), p3(x3, y3) and p4(x4, y4) in the two-dimensional grid is taken as the corresponding grid range of the other vehicle in the two-dimensional grid, the quadrilateral range composed of min_indexX, max_indexX, min_indexY and max_indexY in the two-dimensional grid is taken as the corresponding grid range of the four extreme points in the two-dimensional grid, and the corresponding grid range of the other vehicle in the two-dimensional grid is contained in the corresponding grid range of the four extreme points in the two-dimensional grid. The corresponding grid range of the four extreme points in the two-dimensional grid is taken as the screening range. Based on this, in determining the non-ground points located in the corresponding range of the other vehicle in the vehicle coordinate system, only the cells belonging to the screening range need to be traversed.

[0153] In the traversal, it is judged by the cross multiplication method whether the position of the two-dimensional grid index corresponding to each cell in the screening range is within the grid range formed by the two-dimensional grid index coordinates corresponding to the four corner points of the other vehicle. If it is, the cell corresponding to the index point (x, y) is marked with the vehicle attribute.

[0154] The cross multiplication calculation method is as follows:

[0155] The cross multiplication results of each edge of the quadrilateral formed by the two-dimensional grid index point of the cell and the two-dimensional grid index coordinates corresponding to the four corner points of the other vehicle are calculated: d1 = point × (p1-p2) d2 = point × (p2-p3) d3 = point × (p3-p4) d4 = point × (p4-p1)

[0156] It is checked whether the signs of the cross multiplication results (d1, d2, d3, d4) are consistent: if all the results are greater than or equal to 0, or all the results are less than or equal to 0, the cell point is within the grid range corresponding to the other vehicle. Otherwise, the cell point is not within the grid range corresponding to the other vehicle.

[0157] Correspondingly, in the above step S25, based on the adjusted attribute marking of the cell, the non-ground point cloud is divided into obstacle point cloud and noise point cloud, and further comprising:

[0158] Before judging whether the non-ground point is mapped to the obstacle cell marked with the obstacle attribute or to the noise cell marked with the noise attribute, it is judged whether the cell to which the non-ground point is mapped is marked with the vehicle attribute;

[0159] In the case where the cell to which the non-ground point is mapped is marked with the vehicle attribute, the non-ground point is divided into obstacle point cloud;

[0160] In the case where the cell to which the non-ground point is mapped is not marked with the vehicle attribute, the step of judging whether the non-ground point is mapped to the obstacle cell marked with the obstacle attribute or to the noise cell marked with the noise attribute is performed.

[0161] In some embodiments, the point cloud mapped in the cell marked with the vehicle attribute is the point cloud corresponding to the other vehicle near the host vehicle, and the other vehicle is a real existing obstacle, so the non-ground point cloud mapped in the cell marked with the vehicle attribute can be directly divided into obstacle point cloud. In this way, when the point cloud is divided based on the obstacle attribute or the noise attribute, it can be avoided to be divided incorrectly, thereby improving the accuracy of point cloud division.

[0162] In some embodiments, in order to improve the accuracy of point cloud division, in the above step S24, based on the non-ground points mapped into the cell, the attribute label is added to the cell, which can also include:

[0163] Obtaining target obstacle information detected by the laser radar neural network model, the target obstacle information being used to indicate a target bounding box corresponding to the target obstacle;

[0164] Determining whether the non-ground points mapped into the cell contain non-ground points located in the target bounding box;

[0165] In the case that the non-ground points mapped into the cell contain non-ground points located in the target bounding box, the cell is marked as a laser network detection attribute.

[0166] Here, the laser radar neural network model is a neural network model used for detecting the target obstacle based on the point cloud information collected by the laser radar, wherein the target obstacle detected by the laser radar neural network model is a real existing obstacle. Based on this, when determining whether the non-ground points mapped into the cell contain non-ground points located in the target bounding box,

[0167] The horizontal spatial position corresponding to the target obstacle in the vehicle coordinate system can be determined based on the obtained target obstacle information, that is, the range corresponding to the target detection box in the vehicle coordinate system, and then it is judged whether the two-dimensional coordinates of the non-ground points in the vehicle coordinate system are located in the range corresponding to the target detection box in the vehicle coordinate system. If yes, it is determined that the non-ground points are located in the target bounding box, otherwise it is determined that the non-ground points are not located in the target bounding box. In this way, it can be determined whether the non-ground points mapped into the cell contain non-ground points located in the target bounding box, and further in the case that the non-ground points mapped into the cell contain non-ground points located in the target bounding box, the cell is marked as a laser network detection attribute.

[0168] In some embodiments of the present application, the target obstacle information usually contains the rotation angle of the target detection box rotation, the two-dimensional coordinate position of the center point of the target detection box in the vehicle coordinate system box. center. x and box. center. y, and the length box. length and width box. width of the target detection box. Based on this, whether the non-ground points are located inside the target detection box can be determined by the following way:

[0169] The distance relative_x of the non-ground point to the center point of the target box in the x direction and the distance relative_y of the non-ground point to the center point of the target box in the y direction are calculated by the following formulas: relative_x = (x - box.center.x) * cos_rotation - (y - box.center.y) * sin_rotation relative_y = (x - box.center.x) * sin_rotation + (y - box.center.y) * cos_rotation

[0170] In the formula, cos_rotation represents the cosine value of the rotation angle of the target detection box, and sin_rotation represents the sine value of the rotation angle of the target detection box.

[0171] By the above formula, the horizontal and vertical distances of the non-ground point to the center point of the target detection box are calculated. If the horizontal distance relative_x is within half of the length of the target detection box and the vertical distance relative_y is within half of the width of the target detection box, it indicates that the x value of the coordinates of the non-ground point in the vehicle coordinate system is within the range of -box.length / 2 and box.length / 2, and the y value of the non-ground point in the vehicle coordinate system is within the range of -box.width / 2 and box.width / 2. At this time, it is determined that the non-ground point is located inside the target detection box.

[0172] In this way, by judging whether the coordinates of the non-ground point in the vehicle coordinate system are within the range of the rectangular box, it can be accurately determined whether the non-ground point is located inside the rectangular box.

[0173] Correspondingly, in the above step S25, based on the adjusted attribute label of the cell, the non-ground point cloud is divided into an obstacle point cloud and a noise point cloud, and further comprising:

[0174] Before determining whether the non-ground point is mapped to an obstacle cell labeled as an obstacle attribute or a noise cell labeled as a noise attribute, it is determined whether the cell to which the non-ground point is mapped is labeled as a laser network detection attribute;

[0175] In the case where the cell to which the non-ground point is mapped is labeled as a laser network detection attribute, the non-ground point is divided into an obstacle point cloud;

[0176] In the case where the cell to which the non-ground point is mapped is not labeled as a laser network detection attribute, the step of determining whether the non-ground point is mapped to an obstacle cell labeled as an obstacle attribute or a noise cell labeled as a noise attribute is performed.

[0177] In some embodiments, the point cloud mapped in the cell added with the laser network detection attribute label is the point cloud corresponding to the target obstacle obtained through model detection, and the target obstacle is a real obstacle, so the non-ground point cloud mapped in the cell labeled as the laser network detection attribute can be directly divided into obstacle point cloud. In this way, when the point cloud is divided based on the obstacle attribute or the noise attribute, the point cloud can be divided correctly, thereby improving the accuracy of the point cloud division.

[0178] In some embodiments, in order to improve the accuracy of the point cloud division, in the step S24, the attribute label is added to the cell based on the non-ground point mapped into the cell, and the step can further include:

[0179] Obtaining point cloud information detected by a millimeter wave radar in the vehicle, and the non-ground point cloud information includes dynamic obstacle point cloud and static obstacle point cloud;

[0180] Mapping the dynamic obstacle point cloud and the static obstacle point cloud into a two-dimensional grid;

[0181] In the case that the dynamic obstacle point cloud is mapped into the cell, labeling the cell as a millimeter wave dynamic attribute;

[0182] In the case that the static obstacle point cloud is mapped into the cell, labeling the cell as a millimeter wave static attribute. Similarly, the dynamic obstacle point cloud and the static obstacle point cloud can also be mapped into the two-dimensional grid by the following formula: indexX=floor((x-rxMin) / cellSize) indexY=floor((y-ryMin) / cellSize)

[0183] Correspondingly, in the step S25, based on the adjusted attribute label of the cell, the non-ground point cloud is divided into obstacle point cloud and noise point cloud, and the step can further include:

[0184] Before judging whether the non-ground point is mapped into the obstacle cell labeled as the obstacle attribute or mapped into the noise cell labeled as the noise attribute, judging whether the cell to which the non-ground point is mapped is labeled as the millimeter wave dynamic attribute or the millimeter wave static attribute;

[0185] In the case that the cell to which the non-ground point is mapped is labeled as the millimeter wave dynamic attribute, dividing the non-ground point into obstacle point cloud;

[0186] In the case that the cell to which the non-ground point is mapped is labeled as the millimeter wave static attribute, if the number of non-ground point clouds in the cell to which the non-ground point is mapped is greater than a ninth number threshold, dividing the non-ground point into obstacle point cloud;

[0187] In the case that the cell to which the non-ground point is mapped is not marked with the millimeter wave dynamic attribute and the millimeter wave static attribute, or the cell to which the non-ground point is mapped is marked as the millimeter wave static attribute, but the number of the non-ground point cloud in the cell to which the non-ground point is mapped is less than or equal to the ninth number threshold, the step of judging whether the non-ground point is mapped to the obstacle cell marked as the obstacle attribute or the noise cell marked as the noise attribute is performed.

[0188] In some embodiments, the point cloud partitioning can be assisted based on the millimeter wave radar-based obstacle detection result, which can improve the point cloud partitioning accuracy.

[0189] In some embodiments, whether the scene in front of the vehicle is in a rain, fog, snow or dust environment can also be determined based on the attribute marked by the cell, and the judgment method is as follows:

[0190] If the point cloud information collected by the laser radar in the vehicle satisfies the following conditions for N consecutive frames, it is determined that the front of the vehicle is in a rain, fog, snow or dust environment:

[0191] The number of cells marked as the noise attribute in the two-dimensional grid to which the non-ground point cloud collected in the preset distance in front of the vehicle for more than two consecutive frames is mapped is greater than a certain threshold, and the average intensity of the non-ground point cloud in the cell marked as the noise attribute is greater than a certain threshold.

[0192] If the point cloud information collected by the laser radar in the vehicle does not satisfy the following conditions for M consecutive frames, it is determined that the front of the vehicle is in a normal environment:

[0193] The number of cells marked as the noise attribute in the two-dimensional grid to which the non-ground point cloud collected in the preset distance in front of the vehicle for more than two consecutive frames is mapped is greater than a certain threshold, and the average intensity of the non-ground point cloud in the cell marked as the noise attribute is greater than a certain threshold.

[0194] The values of each threshold, the preset distance, and N and M used above can be set according to actual conditions.

[0195] Further, in the case that it is determined that the front of the vehicle is in a rain, fog, snow or dust environment, whether it is a rain, fog, snow or dust environment can be further determined by the following method:

[0196] Obtain the ground point cloud collected by the laser radar in the vehicle;

[0197] Map the ground point cloud to the two-dimensional grid corresponding to the detection range;

[0198] Determine the average intensity of the ground point cloud mapped to the two-dimensional grid;

[0199] If the average intensity is greater than a certain threshold, it is determined that the front of the vehicle is a dust environment, otherwise it is determined that the front of the vehicle is a rain, fog or snow environment.

[0200] This is because the reflection intensity of dust is greater than that of rain, fog or snow.

[0201] The threshold value can also be set according to actual conditions.

[0202] Based on the obstacle detection denoising method provided in the above embodiments, the application also provides a specific implementation of the obstacle detection denoising device. Please refer to the following embodiments.

[0203] Referring to FIG. 3, it is a schematic diagram of the obstacle detection denoising device provided by the embodiments of the application, as shown in FIG. 3, the device 300 can include the following modules:

[0204] The acquisition module 301 is configured to acquire a non-ground point cloud collected by a vehicle;

[0205] The grid construction module 302 is configured to construct a two-dimensional grid corresponding to a detection range based on a vehicle coordinate system of the vehicle and a grid resolution, and the two-dimensional grid includes a plurality of cells;

[0206] The mapping module 303 is configured to map the non-ground point cloud to the two-dimensional grid;

[0207] The attribute marking module 304 is configured to, for each cell in the two-dimensional grid that has a non-ground point mapped therein, add an attribute mark to the cell based on the non-ground point mapped to the cell, and the attribute marked by the cell at least includes one of an obstacle attribute and a noise attribute, wherein the obstacle attribute is used to indicate that the non-ground point mapped to the cell is an obstacle point, and the noise attribute is used to indicate that the non-ground point mapped to the cell is a noise point;

[0208] The point cloud division module 305 is configured to divide the non-ground point cloud into an obstacle point cloud and a noise point cloud based on the attribute marked by each cell in the two-dimensional grid.

[0209] The obstacle detection denoising device provided by the embodiment of the present application obtains a non-ground point cloud collected by a vehicle; a two-dimensional grid corresponding to a detection range is constructed based on a vehicle coordinate system of the vehicle and a grid resolution, the two-dimensional grid includes a plurality of cells, and the non-ground point cloud is mapped to the two-dimensional grid; for each cell in the two-dimensional grid that has a non-ground point mapped thereto, an attribute label is added to the cell based on the non-ground point mapped to the cell, and the attribute label of the cell at least includes one of an obstacle attribute and a noise attribute; and the non-ground point cloud is divided into an obstacle point cloud and a noise point cloud based on the attribute labels of the cells in the two-dimensional grid. According to the embodiment of the present application, the non-ground point cloud can be divided into the obstacle point cloud and the noise point cloud, so that obstacle detection can be performed based on the obstacle point cloud, and thus the influence of the noise point cloud on the detection result can be reduced, and the obstacle detection accuracy can be improved.

[0210] In some embodiments, the attribute labeling module 304 is configured to:

[0211] filtering, from the non-ground points mapped to the cell, a low-intensity point;

[0212] determining a ratio of a number of the low-intensity points to a number of the non-ground points mapped to the cell;

[0213] in a case where the ratio is greater than a ratio threshold, labeling the cell as the noise attribute;

[0214] in a case where the ratio is less than or equal to the ratio threshold, labeling the cell as the obstacle attribute.

[0215] In some embodiments, the attribute labeling module 304 is configured to:

[0216] for each non-ground point mapped to the cell, in a case where a horizontal radial distance of the non-ground point does not belong to a set dynamic intensity radial threshold filtering range, setting a set minimum intensity threshold as an intensity threshold corresponding to the non-ground point, and in a case where the horizontal radial distance of the non-ground point belongs to the set dynamic intensity radial threshold filtering range, determining the intensity threshold corresponding to the non-ground point based on the horizontal radial distance of the non-ground point, the minimum intensity threshold, the dynamic intensity radial threshold filtering range, and a set maximum intensity threshold;

[0217] in a case where an intensity of the non-ground point is less than the intensity threshold corresponding to the non-ground point, determining that the non-ground point is the low-intensity point.

[0218] In some embodiments, the attribute of the cell label further comprises one of a high attribute and a low attribute, and one of a far distance attribute and a near distance attribute, wherein the high attribute is used to indicate that there is a non-ground point with a height greater than a first height threshold in the non-ground points mapped into the cell, the low attribute is used to indicate that the heights of the non-ground points mapped into the cell are all less than or equal to the first height threshold, the far distance attribute is used to indicate that the minimum height of the non-ground points mapped into the cell is greater than a second height threshold, and the near distance attribute is used to indicate that the minimum height of the non-ground points mapped into the cell is less than or equal to the second height threshold;

[0219] The device 300 further comprises an adjusting module, configured to:

[0220] In a case where the cell label is the noise attribute and the low attribute, the first neighborhood cells located within a first neighborhood radius of the cell are traversed;

[0221] In a case where, in the first neighborhood cells, the number of the first neighborhood cells labeled as the noise attribute is less than a first number threshold, the noise attribute of the cell is updated to the obstacle attribute;

[0222] In a case where the cell label is the far distance attribute and the number of the non-ground point clouds mapped into the cell is greater than a second number threshold, the second neighborhood cells located within a second neighborhood radius of the cell are traversed;

[0223] In a case where, in the second neighborhood cells, the number of the second neighborhood cells labeled as the near distance attribute is greater than a third number threshold, the far distance attribute of the cell is updated to the near distance attribute;

[0224] The third neighborhood cells located within a third neighborhood radius outside the vehicle body contour of the vehicle are traversed;

[0225] In a case where, in the third neighborhood cells, there is a third neighborhood cell labeled as the obstacle attribute, the fourth neighborhood cells located within a fourth neighborhood radius of the third neighborhood cell labeled as the obstacle attribute are traversed;

[0226] In a case where, in the fourth neighborhood cells, the number of the fourth neighborhood cells labeled as the noise attribute is greater than a fourth number threshold, the obstacle attribute of the third neighborhood cell labeled as the obstacle attribute in the third neighborhood cells is updated to the noise attribute;

[0227] The fifth neighborhood cells located within a fifth neighborhood radius outside the vehicle body contour of the vehicle are traversed;

[0228] In the case that there is a fifth-neighbor cell labeled with an obstacle attribute, if the number of non-ground point clouds mapped into the fifth-neighbor cell labeled with the obstacle attribute is less than a fifth number threshold, and the minimum height of the non-ground point clouds mapped into the fifth-neighbor cell labeled with the obstacle attribute is greater than a third height threshold, the obstacle attribute of the fifth-neighbor cell labeled with the obstacle attribute is updated to a noise attribute.

[0229] In some embodiments, the point cloud division module 305 is configured to:

[0230] For each non-ground point in each non-ground point cloud, the following steps are performed respectively:

[0231] Determine whether the non-ground point is mapped into a two-dimensional grid;

[0232] In the case that the non-ground point is not mapped into the two-dimensional grid, divide the non-ground point into an obstacle point cloud;

[0233] In the case that the non-ground point is mapped into the two-dimensional grid, determine whether the non-ground point is mapped into an obstacle cell labeled with an obstacle attribute, or a noise cell labeled with a noise attribute;

[0234] In the case that the non-ground point is mapped into the obstacle cell, if the obstacle cell is labeled with a far distance attribute, and the average intensity of the non-ground points mapped into the obstacle cell is less than a first intensity threshold, divide the non-ground point into a noise point cloud;

[0235] If the minimum height of the obstacle cell from the ground is greater than a fourth height threshold, and the average intensity of the non-ground points mapped into the obstacle cell is less than the first intensity threshold, divide the non-ground point into a noise point cloud, the minimum height of the cell from the ground being the minimum height of the non-ground points mapped into the cell from the ground;

[0236] If the minimum height of the obstacle cell from the ground is greater than the fourth height threshold, and the number of non-ground points mapped into the obstacle cell is less than a sixth number threshold, divide the non-ground point into a noise point cloud;

[0237] Otherwise, divide the non-ground point into an obstacle point cloud;

[0238] In the case that the non-ground point is mapped into the noise cell, if the minimum height of the noise cell from the ground is less than a fifth height threshold, and the number of non-ground points mapped into the noise cell is greater than a seventh number threshold, divide the non-ground point into an obstacle point cloud, otherwise, divide the non-ground point into a noise point cloud.

[0239] In some embodiments, the attribute labeling module 304 is further configured to:

[0240] cluster the non-ground point cloud to obtain a plurality of point cloud clusters;

[0241] perform linear fitting on the point cloud in each point cloud cluster respectively to obtain a set of line segments corresponding to each point cloud cluster;

[0242] take the point cloud cluster containing a line with a length greater than a length threshold in the corresponding set of line segments as a point cloud cluster corresponding to a real obstacle;

[0243] if the non-ground points mapped into the cell contain points in the point cloud cluster corresponding to the real obstacle, mark the cell as a regular attribute, and the regular attribute is used to indicate that the non-ground points mapped into the cell contain points corresponding to the real obstacle;

[0244] The point cloud division module 305 is further configured to:

[0245] before determining whether the non-ground point is mapped into the obstacle cell marked as the obstacle attribute or the noise cell marked as the noise attribute, determine whether the cell to which the non-ground point is mapped is marked as the regular attribute;

[0246] if the number of non-ground point clouds contained in the cell to which the non-ground point is mapped is greater than an eighth number threshold, mark the point as an obstacle point cloud in a case where the cell to which the non-ground point is mapped is marked as the regular attribute;

[0247] in a case where the cell to which the non-ground point is mapped is not marked as the regular attribute, perform the step of determining whether the non-ground point is mapped into the obstacle cell marked as the obstacle attribute or the noise cell marked as the noise attribute.

[0248] In some embodiments, the attribute marking module 304 is further configured to:

[0249] obtain vehicle on-board unit information of the vehicle, the vehicle on-board unit information including related information of other vehicles, the other vehicles being vehicles capable of communicating with the vehicle on-board unit of the vehicle;

[0250] determine a corresponding range of the other vehicles in a vehicle coordinate system of the vehicle based on the vehicle on-board unit information;

[0251] determine whether the cell is located within the corresponding range of the other vehicles in the vehicle coordinate system;

[0252] in a case where the cell is located within the corresponding range of the other vehicles in the vehicle coordinate system, mark the cell as a vehicle attribute;

[0253] The point cloud division module 305 is configured to:

[0254] before determining whether the non-ground point is mapped to a barrier cell marked as the barrier attribute or a noise cell marked as the noise attribute, determining whether the cell to which the non-ground point is mapped is marked as the vehicle attribute;

[0255] in a case where the cell to which the non-ground point is mapped is marked as the vehicle attribute, dividing the non-ground point into the barrier point cloud;

[0256] in a case where the cell to which the non-ground point is mapped is not marked as the vehicle attribute, performing the step of determining whether the non-ground point is mapped to a barrier cell marked as the barrier attribute or a noise cell marked as the noise attribute.

[0257] In some embodiments, the attribute marking module 304 is configured to:

[0258] obtain target barrier information detected by the LiDAR neural network model, the target barrier information being used to indicate a target bounding box corresponding to a target barrier;

[0259] determine whether the non-ground point located in the target bounding box is contained in the non-ground points mapped into the cell;

[0260] in a case where the non-ground point located in the target bounding box is contained in the non-ground points mapped into the cell, marking the cell as the LiDAR network detection attribute;

[0261] The point cloud dividing module 305 is configured to:

[0262] before determining whether the non-ground point is mapped to a barrier cell marked as the barrier attribute or a noise cell marked as the noise attribute, determining whether the cell to which the non-ground point is mapped is marked as the LiDAR network detection attribute;

[0263] in a case where the cell to which the non-ground point is mapped is marked as the LiDAR network detection attribute, dividing the non-ground point into the barrier point cloud;

[0264] in a case where the cell to which the non-ground point is mapped is not marked as the LiDAR network detection attribute, performing the step of determining whether the non-ground point is mapped to a barrier cell marked as the barrier attribute or a noise cell marked as the noise attribute.

[0265] In some embodiments, the attribute marking module 304 is further configured to:

[0266] obtain point cloud information detected by the millimeter wave radar in the vehicle, the non-ground point cloud information including dynamic barrier point cloud and static barrier point cloud;

[0267] mapping the dynamic barrier point cloud and the static barrier point cloud into a two-dimensional grid;

[0268] in case that a dynamic obstacle point cloud is mapped into the cell, marking the cell as a millimeter wave dynamic attribute;

[0269] in case that a static obstacle point cloud is mapped into the cell, marking the cell as a millimeter wave static attribute;

[0270] The point cloud division module 305 is configured to:

[0271] before determining whether the non-ground point is mapped into the obstacle cell marked as the obstacle attribute or into the noise cell marked as the noise attribute, determining whether the cell to which the non-ground point is mapped is marked as the millimeter wave dynamic attribute or the millimeter wave static attribute;

[0272] in case that the cell to which the non-ground point is mapped is marked as the millimeter wave dynamic attribute, dividing the non-ground point into the obstacle point cloud;

[0273] in case that the cell to which the non-ground point is mapped is marked as the millimeter wave static attribute, if the number of non-ground point clouds in the cell to which the non-ground point is mapped is greater than the ninth quantity threshold, dividing the non-ground point into the obstacle point cloud;

[0274] in case that the cell to which the non-ground point is mapped is not marked as the millimeter wave dynamic attribute or the millimeter wave static attribute, or in case that the cell to which the non-ground point is mapped is marked as the millimeter wave static attribute but the number of non-ground point clouds in the cell to which the non-ground point is mapped is less than or equal to the ninth quantity threshold, performing the step of determining whether the non-ground point is mapped into the obstacle cell marked as the obstacle attribute or into the noise cell marked as the noise attribute.

[0275] The obstacle detection and denoising apparatus provided by the embodiments of the present application can implement each process implemented by the method embodiment of FIG. 2, and thus details are not repeated here.

[0276] FIG. 4 shows a hardware structure schematic diagram of an electronic device provided by an embodiment of the present application.

[0277] The electronic device 400 can include a processor 401 and a memory 402 storing computer program instructions.

[0278] Specifically, the processor 401 described above can include a central processing unit (CPU), or an application specific integrated circuit (ASIC), or can be configured as one or more integrated circuits implementing the embodiments of the present application.

[0279] The memory 402 can include mass storage for data or instructions. As an example and not by way of limitation, the memory 402 can include a hard disk drive (HDD), a floppy disk drive, flash memory, an optical disc (e.g., a compact disc (CD) or a digital versatile disc (DVD)), a solid-state drive (SSD), a USB drive, or a combination of two or more of these. Where appropriate, the memory 402 can include removable or non-removable (or fixed) media. Where appropriate, the memory 402 can be internal or external to the integrated gateway disaster recovery device. In particular embodiments, the memory 402 is non-volatile, solid-state memory.

[0280] The memory 402 can include read-only memory (ROM), random-access memory (RAM), a magnetic disk storage medium, an optical storage medium, flash memory devices, electrical, optical, or other physical / tangible memory storage devices. Thus, in general, the memory 402 includes one or more tangible (non-transitory) computer-readable storage media (e.g., a memory device) encoded with software that, when executed (by one or more processors), is capable of operating as described above to perform the operations described above in connection with any of the embodiments of the obstacle detection denoising method.

[0281] The processor 401 implements the obstacle detection denoising method of any of the above embodiments by reading and executing computer program instructions stored in the memory 402.

[0282] In one example, the electronic device 400 also includes a communication interface 403 and a bus 410. As shown in FIG. 4, the processor 401, the memory 402, and the communication interface 403 are connected by the bus 410 and complete communication with each other.

[0283] The communication interface 403 is mainly used to realize the communication between the modules, devices, units and / or equipment in the embodiments of the present application.

[0284] Bus 410 includes a hardware, software, or both that couples components of the online data traffic metering device to each other. As an example without limitation, bus can include an Accelerated Graphics Port (AGP) or other graphics bus, an Enhanced Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), a HyperTransport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an InfiniBand (IB) interconnect, a Low Pin Count (LPC) bus, a memory bus, a Micro Channel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local (VLB) bus, or another suitable bus or combination of busses, or two or more of these busses. Where appropriate, bus 410 can include one or more buses. Although this application describes and illustrates a particular bus, this application contemplates any suitable bus or interconnect.

[0285] In addition, in combination with the obstacle detection denoising method in the above embodiments, the embodiments of the present application can provide a computer storage medium to implement. The computer storage medium has computer program instructions stored thereon; the computer program instructions are executed by a processor to implement any one of the obstacle detection denoising methods in the above embodiments.

[0286] The embodiments of the present application also provide a computer program product, comprising a computer program, the computer program is executed by a processor to implement any one of the obstacle detection denoising methods in the above embodiments.

[0287] It needs to be clear that the present application is not limited to the specific configurations and processes described above and shown in the drawings. For the sake of brevity, detailed descriptions of well-known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present application is not limited to the specific steps described and shown, and those skilled in the art can make various changes, modifications and additions, or change the order between steps, after understanding the spirit of the present application.

[0288] The functions noted in the description of the structural block diagrams above can be implemented as hardware, software, firmware, or a combination thereof. When implemented in hardware, they can be, for example, electronic circuits, application specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, and the like. When implemented in software, the elements of the present application are program or code segments that are used to perform the required tasks. The program or code segments can be stored in a machine-readable medium, or transmitted through a data signal carried in a carrier wave over a transmission medium or communication link. A "machine-readable medium" includes any medium that can store or transport information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROM, flash memory, erasable ROM (EROM), floppy disks, CD-ROMs, optical disks, hard disks, fiber optic media, radio frequency (RF) links, and the like. The code segments can be downloaded via computer networks such as the Internet, intranets, and the like.

[0289] It is also important to note that the examples mentioned in the present application describe some methods or systems based on a series of steps or devices. However, the present application is not limited to the order of the steps mentioned above, that is, the steps can be performed in the order mentioned in the examples, or in a different order from the examples, or several steps can be performed simultaneously.

[0290] The computer program instructions can also be loaded onto a computer, other programmable data processing apparatus, or other processing device to cause a series of operational steps to be performed on the computer, other programmable apparatus or other processing device to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide processes for implementing the functions / acts specified in the flowchart and / or block diagram block or blocks. These computer program instructions can also be stored in a computer readable medium that can direct a computer, other programmable data processing apparatus, or other processing device to operate in a particular manner, such that the instructions stored in the computer readable medium produce an article of manufacture including instructions which implement the function / act specified in the flowchart and / or block diagram block or blocks. The computer program instructions can also be loaded onto a computer, other programmable data processing apparatus, or other processing device to cause a series of operational steps to be performed on the computer, other programmable apparatus or other processing device to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide processes for implementing the functions / acts specified in the flowchart and / or block diagram block or blocks. These computer program instructions can also be stored in a computer readable medium that can direct a computer, other programmable data processing apparatus, or other processing device to operate in a particular manner, such that the instructions stored in the computer readable medium produce an article of manufacture including instructions which implement the function / act specified in the flowchart and / or block diagram block or blocks.

[0291] The above merely describes a specific implementation of the present application. Those skilled in the art can clearly understand the specific working processes of the system, modules and units described above for the convenience and brevity of description, and can refer to the corresponding processes in the foregoing method embodiments, which will not be described herein again. It should be understood that the protection scope of the present application is not limited to this, and any person skilled in the art can easily think of various equivalent modifications or replacements within the technical range disclosed by the present application, and these modifications or replacements should be covered within the protection scope of the present application.

Claims

1. A method for obstacle detection denoising, comprising: obtaining a non-ground point cloud collected by a vehicle; constructing a two-dimensional grid corresponding to a detection range based on a vehicle coordinate system of the vehicle and a grid resolution, the two-dimensional grid comprising a plurality of cells; mapping the non-ground point cloud to the two-dimensional grid; for each cell in the two-dimensional grid having a non-ground point mapped thereto, adding an attribute label to the cell based on the non-ground points mapped to the cell, the attribute label of the cell comprising at least one of an obstacle attribute and a noise attribute, wherein the obstacle attribute indicates that the non-ground points mapped to the cell are obstacle points, and the noise attribute indicates that the non-ground points mapped to the cell are noise points; dividing the non-ground point cloud into an obstacle point cloud and a noise point cloud based on the attribute labels of the cells in the two-dimensional grid.

2. The method of claim 1, wherein, The adding of the attribute label to the cell based on the non-ground points mapped to the cell comprises: screening low-intensity points from the non-ground points mapped to the cell; determining a ratio of a number of the low-intensity points to a number of the non-ground points mapped to the cell; in a case where the ratio is greater than a proportion threshold, labeling the cell as the noise attribute; in a case where the ratio is less than or equal to the proportion threshold, labeling the cell as the obstacle attribute.

3. The method of claim 2, wherein, The screening of the low-intensity points from the non-ground points mapped to the cell comprises: for each non-ground point mapped to the cell, in a case where a horizontal radial distance of the non-ground point does not belong to a set dynamic intensity radial threshold filtering range, setting a set minimum intensity threshold as an intensity threshold corresponding to the non-ground point, and in a case where the horizontal radial distance of the non-ground point belongs to the set dynamic intensity radial threshold filtering range, determining the intensity threshold corresponding to the non-ground point based on the horizontal radial distance of the non-ground point, the minimum intensity threshold, the dynamic intensity radial threshold filtering range, and a set maximum intensity threshold; in a case where an intensity of the non-ground point is less than the intensity threshold corresponding to the non-ground point, determining the non-ground point as a low-intensity point.

4. The method of any one of claims 1-3, wherein, The attribute label of the cell further comprises one of a high attribute and a low attribute, wherein the high attribute indicates that there is a non-ground point with a height greater than a first height threshold among the non-ground points mapped to the cell, and the low attribute indicates that heights of the non-ground points mapped to the cell are all less than or equal to the first height threshold. Before the dividing of the non-ground point cloud into the obstacle point cloud and the noise point cloud based on the attribute labels of the cells in the two-dimensional grid, the method further comprises: in a case where the cell is labeled as the noise attribute and the low attribute, traversing first neighborhood cells located within a first neighborhood radius around the cell; in a case where a number of the first neighborhood cells labeled as the noise attribute is less than a first number threshold among the first neighborhood cells, updating the noise attribute of the cell to the obstacle attribute.

5. The method of any one of claims 1-4, wherein, The attribute of the cell label further comprises one of a long-distance attribute and a short-distance attribute, wherein the long-distance attribute is used to indicate that a minimum height of a non-ground point mapped into the cell is greater than a second height threshold, and the short-distance attribute is used to indicate that the minimum height of the non-ground point mapped into the cell is less than or equal to the second height threshold; Before the non-ground point cloud is divided into the obstacle point cloud and the noise point cloud based on the attribute of each cell label in the two-dimensional grid, the method further comprises: In a case where the cell label is the long-distance attribute and a number of non-ground point clouds mapped into the cell is greater than a second number threshold, a second neighborhood cell located within a second neighborhood radius of a periphery of the cell is traversed; In a case where a number of second neighborhood cells labeled as the short-distance attribute in the second neighborhood cell is greater than a third number threshold, the long-distance attribute of the cell is updated to the short-distance attribute.

6. The method of any one of claims 1-5, wherein, Before the non-ground point cloud is divided into the obstacle point cloud and the noise point cloud based on the attribute of each cell label in the two-dimensional grid, the method further comprises: A third neighborhood cell located within a third neighborhood radius outside a body contour of the vehicle is traversed; In a case where there is a third neighborhood cell labeled as the obstacle attribute in the third neighborhood cell, a fourth neighborhood cell located within a fourth neighborhood radius of the third neighborhood cell labeled as the obstacle attribute is traversed; In a case where a number of fourth neighborhood cells labeled as the noise attribute in the fourth neighborhood cell is greater than a fourth number threshold, an obstacle attribute of the third neighborhood cell labeled as the obstacle attribute in the third neighborhood cell is updated to the noise attribute; A fifth neighborhood cell located within a fifth neighborhood radius outside the body contour of the vehicle is traversed; In a case where there is a fifth neighborhood cell labeled as the obstacle attribute in the fifth neighborhood cell, if a number of non-ground point clouds mapped into the fifth neighborhood cell labeled as the obstacle attribute is less than a fifth number threshold and a minimum height of the non-ground point clouds mapped into the fifth neighborhood cell labeled as the obstacle attribute is greater than a third height threshold, an obstacle attribute of the fifth neighborhood cell labeled as the obstacle attribute is updated to the noise attribute.

7. The method of any one of claims 1-6, wherein, The dividing of the non-ground point cloud into the obstacle point cloud and the noise point cloud based on the attribute of each cell label in the two-dimensional grid comprises: For each non-ground point in each non-ground point cloud, the following steps are respectively performed: It is determined whether the non-ground point is mapped into the two-dimensional grid; In a case where the non-ground point is not mapped into the two-dimensional grid, the non-ground point is divided into the obstacle point cloud; In a case where the non-ground point is mapped into the two-dimensional grid, it is determined whether the non-ground point is mapped into an obstacle cell labeled as the obstacle attribute or a noise cell labeled as the noise attribute; In the case that the non-ground point is mapped to the obstacle cell, if the obstacle cell is marked as a long-distance attribute, and an average intensity of the non-ground points mapped to the obstacle cell is less than a first intensity threshold, the non-ground points are divided into a noise point cloud, and the long-distance attribute is used to indicate that a minimum height of the non-ground points mapped to the cell is greater than a second height threshold; if a minimum height of the obstacle cell from the ground is greater than a fourth height threshold, and the average intensity of the non-ground points mapped to the obstacle cell is less than the first intensity threshold, the non-ground points are divided into a noise point cloud, and the minimum height of the cell from the ground is a minimum height of the non-ground points mapped to the cell from the ground, and the intensity of the point cloud is used to indicate a laser reflectivity of an object reflecting the point cloud; if the minimum height of the obstacle cell from the ground is greater than the fourth height threshold, and a number of the non-ground points mapped to the obstacle cell is less than a sixth number threshold, the non-ground points are divided into a noise point cloud; otherwise, the non-ground points are divided into an obstacle point cloud; In the case that the non-ground point is mapped to the noise cell, if a minimum height of the noise cell from the ground is less than a fifth height threshold, and a number of the non-ground points mapped to the noise cell is greater than a seventh number threshold, the non-ground points are divided into an obstacle point cloud, otherwise, the non-ground points are divided into a noise point cloud.

8. The method of claim 7, wherein, The adding of the attribute mark to the cell based on the non-ground point cloud mapped to the cell further comprises: performing clustering processing on the non-ground point cloud to obtain a plurality of point cloud clusters; performing straight line fitting on the point cloud in each point cloud cluster respectively to obtain a straight line segment set corresponding to each point cloud cluster; taking the point cloud cluster containing a straight line with a length greater than a length threshold in the corresponding straight line segment set as a point cloud cluster corresponding to a real obstacle; if the non-ground point mapped to the cell contains a point in the point cloud cluster corresponding to the real obstacle, marking the cell as a regular attribute, and the regular attribute is used to indicate that the non-ground point mapped to the cell contains a point corresponding to a real obstacle; The dividing of the non-ground point cloud into the obstacle point cloud and the noise point cloud based on the adjusted attribute mark of the cell further comprises: before determining whether the non-ground point is mapped to the obstacle cell marked as the obstacle attribute or the noise cell marked as the noise attribute, determining whether the cell to which the non-ground point is mapped is marked as the regular attribute; in the case that the cell to which the non-ground point is mapped is marked as the regular attribute, if a number of the non-ground point cloud contained in the cell to which the non-ground point is mapped is greater than an eighth number threshold, marking the point as the obstacle point cloud; in the case that the cell to which the non-ground point is mapped is not marked as the regular attribute, performing the step of determining whether the non-ground point is mapped to the obstacle cell marked as the obstacle attribute or the noise cell marked as the noise attribute.

9. The method of any one of claims 7-8, wherein, The adding of the attribute mark to the cell based on the non-ground point mapped to the cell further comprises: Obtain vehicle-mounted unit information of the vehicle, the vehicle-mounted unit information comprising relevant information of other vehicles capable of communicating with the vehicle-mounted unit of the vehicle; Determine a corresponding range of the other vehicles in a vehicle coordinate system of the vehicle based on the vehicle-mounted unit information; Determine whether the cell is located within the corresponding range of the other vehicles in the vehicle coordinate system; In the case that the cell is located within the corresponding range of the other vehicles in the vehicle coordinate system, mark the cell as a vehicle attribute; The method further comprises: Before determining whether the non-ground point is mapped to a cell marked as an obstacle attribute or a cell marked as a noise attribute, determining whether the cell to which the non-ground point is mapped is marked as a vehicle attribute; In the case that the cell to which the non-ground point is mapped is marked as a vehicle attribute, dividing the non-ground point into an obstacle point cloud; In the case that the cell to which the non-ground point is mapped is not marked as a vehicle attribute, performing the step of determining whether the non-ground point is mapped to a cell marked as an obstacle attribute or a cell marked as a noise attribute.

10. The method of any one of claims 7-9, wherein, The method further comprises: Obtain target obstacle information detected by a laser radar neural network model, the target obstacle information being used to indicate a target bounding box corresponding to a target obstacle; Determine whether the non-ground point mapped to the cell is located within the target bounding box; In the case that the non-ground point mapped to the cell is located within the target bounding box, mark the cell as a laser network detection attribute; The method further comprises: Before determining whether the non-ground point is mapped to a cell marked as an obstacle attribute or a cell marked as a noise attribute, determining whether the cell to which the non-ground point is mapped is marked as a laser network detection attribute; In the case that the cell to which the non-ground point is mapped is marked as a laser network detection attribute, divide the non-ground point into an obstacle point cloud; In the case that the cell to which the non-ground point is mapped is not marked as a laser network detection attribute, perform the step of determining whether the non-ground point is mapped to a cell marked as an obstacle attribute or a cell marked as a noise attribute.

11. The method of any one of claims 7-10, wherein, The method further comprises: Obtain point cloud information detected by a millimeter wave radar in the vehicle, the non-ground point cloud information comprising dynamic obstacle point cloud and static obstacle point cloud; Map the dynamic obstacle point cloud and the static obstacle point cloud to the two-dimensional grid; in a case where a dynamic obstacle point cloud is mapped into the cell, marking the cell as a millimeter wave dynamic attribute; in a case where a static obstacle point cloud is mapped into the cell, marking the cell as a millimeter wave static attribute; the method further comprises: before determining whether the non-ground point is mapped into an obstacle cell marked as an obstacle attribute or a noise cell marked as a noise attribute, determining whether the cell to which the non-ground point is mapped is marked as a millimeter wave dynamic attribute or a millimeter wave static attribute; in a case where the cell to which the non-ground point is mapped is marked as a millimeter wave dynamic attribute, dividing the non-ground point into an obstacle point cloud; in a case where the cell to which the non-ground point is mapped is marked as a millimeter wave static attribute, if the number of non-ground point clouds in the cell to which the non-ground point is mapped is greater than a ninth quantity threshold, dividing the non-ground point into an obstacle point cloud; in a case where the cell to which the non-ground point is mapped is not marked as a millimeter wave dynamic attribute or a millimeter wave static attribute, or the cell to which the non-ground point is mapped is marked as a millimeter wave static attribute, but the number of non-ground point clouds in the cell to which the non-ground point is mapped is less than or equal to the ninth quantity threshold, performing the step of determining whether the non-ground point is mapped into an obstacle cell marked as an obstacle attribute or a noise cell marked as a noise attribute.

12. An obstacle detection and denoising apparatus, comprising: an acquisition module configured to acquire a non-ground point cloud collected by a vehicle; a grid construction module configured to construct a two-dimensional grid corresponding to a detection range based on a vehicle coordinate system of the vehicle and a grid resolution, the two-dimensional grid comprising a plurality of cells; a mapping module configured to map the non-ground point cloud to the two-dimensional grid; an attribute marking module configured to, for each cell of the two-dimensional grid having a non-ground point mapped thereto, add an attribute mark to the cell based on the non-ground point mapped into the cell, the attribute mark of the cell comprising at least one of an obstacle attribute and a noise attribute, wherein the obstacle attribute is used to indicate that the non-ground point mapped into the cell is an obstacle point, and the noise attribute is used to indicate that the non-ground point mapped into the cell is a noise point; a point cloud division module configured to divide the non-ground point cloud into an obstacle point cloud and a noise point cloud based on the attribute mark of each cell of the two-dimensional grid.

13. An electronic device comprising: a processor and a memory storing computer program instructions; the processor executes the computer program instructions to implement the obstacle detection and denoising method according to any one of claims 1-12.

14. A computer readable storage medium storing computer program instructions, the computer program instructions being executed by a processor to implement the obstacle detection and denoising method according to any one of claims 1-12.

15. A computer program product, instructions in the computer program product, when executed by a processor of an electronic device, cause the electronic device to perform the obstacle detection denoising method of any one of claims 1-12.

Images