Dynamic point cloud detection method and system based on elevation image

Abstract

The invention belongs to the technical field of laser radars. The invention provides a dynamic point cloud detection method and system based on an elevation image. The method comprises the following steps: acquiring ground point cloud, non-ground point cloud and speedometer information of each frame of laser radar point cloud; constructing an elevation image by using the static point cloud and the ground point cloud obtained in the previous frame; carrying out ground continuity restoration on the elevation image; performing shielding detection on the point cloud of the current frame based on an elevation image by combining the odometer information, and determining a dynamic point, a suspected dynamic point and a suspected static point; performing region consistency check on pixels where the dynamic points, the suspected dynamic points and the suspected static points are located, and determining final dynamic points; and performing clustering operation on the detected final dynamic point cloud to obtain a dynamic point cloud detection result. According to the method, the problems of map background point cloud missing, depth image void, easy misjudgment of ground point cloud and the like are effectively solved, so that the accuracy of dynamic object detection is remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of laser radar technology, and in particular to a dynamic point cloud detection method and system based on elevation images. Background Art

[0002] The statements in this section merely provide background art related to the present invention and do not necessarily constitute prior art.

[0003] With the development of unmanned technology, autonomous robots have shown revolutionary potential in many fields, such as driverless taxis, agricultural robots, and factory handling robots, making them increasingly important in daily life. However, when deploying these robots in real environments, detecting and avoiding unknown moving objects becomes a challenge. For example, pedestrians or cyclists who suddenly cross the road may collide with self-driving cars.

[0004] To avoid such accidents, autonomous robots usually rely on the laser radar they carry to detect moving objects. At present, the dynamic object detection method based on laser radar has become a research hotspot, and related research has made important progress. It can be divided into three categories: point cloud segmentation method, ray tracing method and visibility method.

[0005] Traditional point cloud segmentation methods are mainly implemented through clustering. This type of method has a good dynamic object detection effect, but its computational overhead is large. Therefore, semantic segmentation methods based on deep learning have gradually become a hot research object. Although this method is simple and efficient, its model performance is heavily dependent on the quality of the pre-trained dataset and has certain limitations.

[0006] The ray tracing method is based on the principle of ray propagation. It marks the end point of the ray as an occupied voxel through voxel traversal, and the space passed by the ray as a free voxel. Although this method has good theoretical support, it requires the construction of a huge voxel map, which consumes a lot of computing and memory resources. In addition, this method relies on extremely accurate positioning information, which limits its scope of application.

[0007] The visibility method is based on the viewpoint occlusion principle. By constructing a depth image of a map point, it determines whether the current point occludes the map, thereby realizing the detection of dynamic objects. This type of method has the advantages of small computational complexity and high speed. At the same time, it does not require the construction of a large-scale voxel map, so it has a wide range of applications.

[0008] Although visibility methods perform well in dynamic object detection, they have certain limitations: when there is a lack of background map point clouds, it is difficult for such methods to effectively detect dynamic objects, limiting their applicability in some scenarios; due to the lack of continuity of background objects, such methods generally use linear interpolation methods to fill holes in depth images, which also leads to errors; in addition, ground point clouds usually have large incident angles, which makes it easy for such methods to misjudge ground point clouds as dynamic objects. Summary of the invention

[0009] In order to address the deficiencies in the prior art, the present invention provides a dynamic point cloud detection method and system based on elevation images, which effectively overcomes the problems of missing map background point clouds, holes in depth images, and easy misjudgment of ground point clouds, thereby significantly improving the accuracy of dynamic object detection.

[0010] In order to achieve the above object, the present invention adopts the following technical solution: In a first aspect, the present invention provides a dynamic point cloud detection method based on elevation images.

[0011] A dynamic point cloud detection method based on elevation image includes the following processes: Get the ground point cloud, non-ground point cloud and odometer information of each frame of lidar point cloud; Construct an elevation image using the static point cloud and ground point cloud obtained in the previous frame; Perform ground continuity repair on elevation images; In combination with the odometer information, occlusion detection is performed on the point cloud of the current frame based on the elevation image to determine dynamic points, suspected dynamic points, and suspected static points; Performing a regional consistency check on the dynamic point, the suspected dynamic point, and the pixels where the suspected static point is located to determine a final dynamic point; The final detected dynamic point cloud is clustered to obtain the dynamic point cloud detection result.

[0012] In a second aspect, the present invention provides a dynamic point cloud detection system based on elevation images.

[0013] A dynamic point cloud detection system based on elevation image, comprising: The data acquisition unit is configured to: acquire ground point cloud, non-ground point cloud and odometer information of each frame of laser radar point cloud; The elevation image construction unit is configured to: construct an elevation image using a static point cloud and a ground point cloud obtained in a previous frame; The ground continuity repair unit is configured to: perform ground continuity repair on the elevation image; An occlusion detection unit is configured to: perform occlusion detection on the point cloud of the current frame based on the elevation image in combination with the odometer information, and determine dynamic points, suspected dynamic points, and suspected static points; A regional consistency check unit is configured to: perform a regional consistency check on the dynamic point, the suspected dynamic point and the pixel where the suspected static point is located to determine a final dynamic point; The detection result generating unit is configured to: perform a clustering operation on the detected final dynamic point cloud to obtain a dynamic point cloud detection result.

[0014] Compared with the prior art, the present invention has the following beneficial effects: 1. The present invention adopts an elevation image method based on a concentric circle model. By changing the projection direction of the traditional depth image in the visibility method, it avoids the problems of missing map background point cloud and misjudgment of ground point cloud, thereby significantly improving the accuracy of dynamic object detection.

[0015] 2. The present invention proposes a method for ground continuity repair based on the ground point cloud continuity assumption, which effectively fills the hole pixels in the elevation image, increases the ground pixels that can be used for occlusion detection, effectively reduces the misjudgment rate and missed detection rate of dynamic point clouds, and improves the reliability of detection results.

[0016] 3. The present invention introduces a multi-frame occlusion detection and regional consistency check mechanism, which effectively reduces the demand for computing and memory resources, further improves the robustness and real-time performance of dynamic detection, and is suitable for resource-constrained embedded devices and applications in complex dynamic environments.

[0017] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0019] Figure 1 A schematic diagram of a flow chart of a dynamic point cloud detection method based on elevation images provided in Embodiment 1 of the present invention; Figure 2 A schematic diagram of a data acquisition method provided in Example 1 of the present invention; Figure 3 A schematic diagram of an elevation image provided in Example 1 of the present invention; Figure 4 A schematic diagram of a ground continuity repair method provided in Example 1 of the present invention; Figure 5A schematic diagram of the occlusion detection principle provided in Example 1 of the present invention; Figure 6 A schematic diagram of the regional consistency check principle provided in Example 1 of the present invention; Figure 7 A schematic diagram of a dynamic point cloud detection system based on elevation images provided in Example 2 of the present invention. DETAILED DESCRIPTION

[0020] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0021] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0022] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0023] Embodiment 1: This implementation proposes a dynamic point cloud detection method based on elevation images, such as Figure 1 As shown, the following process is included: S1: Obtain the ground point cloud, non-ground point cloud and odometer information of each frame of lidar point cloud; S2: construct an elevation image using the static point cloud and ground point cloud obtained in the previous frame; S3: perform ground continuity repair on elevation images; S4: combining the odometer information, performing occlusion detection on the point cloud of the current frame based on the elevation image, and determining dynamic points, suspected dynamic points, and suspected static points; S5: performing a regional consistency check on the dynamic point, the suspected dynamic point and the pixel where the suspected static point is located to determine a final dynamic point; S6: Perform clustering operation on the final detected dynamic point cloud to obtain the dynamic point cloud detection result.

[0024] In S1 of this implementation, specifically, it includes: Receive the ground point cloud and non-ground point cloud output by the ground point cloud segmentation system, and obtain the odometer data provided by the multi-sensor fusion simultaneous positioning and mapping system to complete the acquisition of input data, such as Figure 2As shown in the figure, the odometer data obtained refers to the relative position and posture between two frames of lidar point clouds. The multi-sensor fusion simultaneous localization and mapping system receives the lidar point cloud and IMU data, outputs the odometer data and point cloud map, outputs the dedistorted point cloud to the ground point cloud segmentation system, and outputs the odometer data to the dynamic point cloud detection system based on elevation images. The ground point cloud segmentation system transmits the ground point cloud and non-ground point cloud to the dynamic point cloud detection system based on elevation images. The dynamic point cloud detection system based on elevation images generates ground point cloud, static point cloud and dynamic point cloud and outputs them to the multi-sensor fusion simultaneous localization and mapping system.

[0025] In S2 of this implementation, specifically, it includes: The elevation image consists of four annular areas, namely annular area 1, annular area 2, annular area 3 and annular area 4. Figure 3 As shown in the figure, the pixel density of these annular areas gradually decreases from the inside to the outside, forming a distribution pattern from dense to sparse. The radial radius formulas of the four annular areas are: (1); Among them, the outer diameter of the annular area is , the inner diameter is ; The outer diameter of the annular region 2 is , the inner diameter is ; The outer diameter of the annular region 3 is , the inner diameter is ; The outer diameter of the annular region 4 is , the inner diameter is , The meaning is meter.

[0026] The pixel resolutions of the four annular areas are: (2); Among them, the radial resolution of the annular region is , the angular resolution is , the number of radial rings in this area is , , the number of sectors in each ring is , ; The radial resolution of annular region 2 is , the angular resolution is , the number of radial rings in this area is , , the number of sectors in each ring is , ; The radial resolution of annular region 3 is , the angular resolution is , the number of radial rings in this area is , , the number of sectors in each ring is , ; The radial resolution of annular region 4 is , the angular resolution is , the number of radial rings in this area is , , the number of sectors in each ring is , .

[0027] The construction of the elevation image is based on the static point cloud and ground point cloud obtained in the previous frame. First, each point in the static point cloud and the ground point cloud is converted into a Cartesian coordinate system. Convert to polar coordinates , the conversion formula is as follows: (3); (4).

[0028] Then, according to the resolution of the elevation image, each point is mapped to the corresponding elevation image pixel. The index of each pixel consists of the ring number and the sector number. , the formula for calculating the pixel index of each point is as follows: (5); In the elevation image, each pixel records the following information: the number of point clouds contained in the pixel, the coordinate values ​​of the point clouds, and the type of point clouds. If the pixel contains a ground point cloud, the pixel is marked as a ground pixel; otherwise, it is a non-ground pixel.

[0029] In S3 of this implementation, based on the ground point cloud continuity assumption, ground continuity restoration is performed on non-ground pixels in the elevation image, specifically, including: S3.1: The repair process starts from the annular area and first finds all the ground pixels in the area. The annular area numbers of the pixels are traversed from large to small, and new ground pixels are searched pixel by pixel along the ray direction toward the center of the elevation image, starting from each ground pixel; If a new ground pixel is found in annular area 1 or annular area 2, all non-ground pixels between the starting point and the new ground pixel are repaired for ground continuity and marked as ground repair pixels; If no new ground pixels are found in the second annular area, the ray traversal is stopped and the ground continuity repair is not performed on the traversed pixels.

[0030] S3.2: After the ground pixel traversal of the annular area 1 is completed, the annular area 2 is processed, all ground pixels in the area are extracted, and the pixel annular area numbers are traversed from large to small. Starting from each ground pixel, new ground pixels are searched pixel by pixel along the ray direction toward the center of the elevation image. If a new ground pixel is found in annular area 2 or annular area 3, all non-ground pixels between the starting point and the new ground pixel are repaired for ground continuity and marked as ground repair pixels; If no new ground pixels are found in the annular area 3, the ray traversal is stopped and the ground continuity repair is not performed on the traversed pixels.

[0031] S3.3: After the ground pixel traversal of the annular area 2 is completed, continue to process the annular area 3. Extract all the ground pixels in the area, traverse from large to small according to the pixel annular area number, take each ground pixel as the starting point, and search for new ground pixels pixel by pixel along the ray direction toward the center of the elevation image; If a new ground pixel is found in annular area 3 or annular area 4, ground continuity repair is performed on all non-ground pixels between the starting point and the new ground pixel, and they are marked as ground repair pixels; If no new ground pixel is found in the annular area 4, the ray traversal is stopped and the ground continuity repair is not performed on the traversed pixels.

[0032] S3.4: After the traversal of the ground pixels in the annular area 3 is completed, the annular area 4 is processed. All ground pixels in the area are extracted, and the pixel annular area numbers are traversed from large to small. Starting from each ground pixel, all non-ground pixels that the ray passes through are repaired along the ray direction to the center of the elevation image, and these pixels are marked as ground repair pixels.

[0033] At this point, the ground continuity restoration of the elevation image is completed, and all ground point clouds that may be missing in non-ground pixels have been supplemented, such as Figure 4 shown.

[0034] Ground continuity repair method: According to the distance between the ground repair pixel and the starting ground pixel and the found ground pixel, the ground point cloud height of the ground repair pixel is calculated using the linear interpolation method. The calculation formula is as follows: (6); Among them, the ground height of the pixel to be repaired is , the ground height of the starting ground pixel is , the ground height of the ground pixel found is , the distance from the pixel to be repaired to the starting ground pixel is , the total distance between the starting ground pixel and the found ground pixel is .

[0035] In S4 of this implementation, specifically, it includes: The elevation images of the last five frames of LiDAR point clouds are always maintained. Therefore, during the system initialization phase, a time window of the first five frames of LiDAR point clouds is required. During this period, all non-ground point clouds are regarded as static point clouds. The point clouds are only used to construct elevation images without occlusion detection.

[0036] Starting from the sixth frame, the point cloud is subjected to occlusion detection. The specific method is as follows: First, the non-ground point cloud of the current frame is transformed into the coordinate system of the previous five frames of elevation images one by one using the odometer data. The transformation formula is as follows: (7); Among them, the current frame lidar point cloud is , the point cloud of the current frame laser radar point cloud is transformed to the point cloud of the previous five frames elevation image coordinate system: , the odometer data of the first five frames is .

[0037] Then, the transformed point cloud is divided into corresponding pixels according to the corresponding elevation image resolution. The division method is: first convert the points from the Cartesian coordinate system to the polar coordinate system, and then map each point to the corresponding pixel according to the resolution of the elevation image.

[0038] Next, occlusion detection is performed on the pixels where the points mapped to the elevation image are located. The principle of occlusion detection is as follows: Figure 5 As shown in the figure, if the difference between the Z-axis height of the point to be detected and the Z-axis height of the highest ground point in the pixel exceeds the set threshold , then it is considered that the point to be detected has blocked the ground point, otherwise, it is considered that there is no occlusion; similarly, if the difference between the Z-axis height of the static point in the pixel and the Z-axis height of the point to be detected exceeds the threshold , then the point to be detected is considered to be blocked by the static point, otherwise, it is considered that there is no occlusion.

[0039] During the occlusion detection process, the following situations may occur: (1) The pixel is a non-ground pixel and there is no point cloud in the pixel: the mapping point is judged as a suspected static point; (2) The pixel is a non-ground pixel and the mapping point occludes the static object in the pixel: the mapping point is judged as a static point; (3) The pixel is a non-ground pixel and the mapping point is blocked by a static object in the pixel: the mapping point is judged as a static point; (4) The pixel is a ground pixel or a ground repair pixel, and the mapping point is blocked by the ground point in the pixel: the mapping point is judged as an invalid point; (5) The pixel is a ground pixel or a ground repair pixel, and the mapping point only occludes the ground point: the mapping point is judged as a dynamic point; (6) The pixel is a ground pixel or a ground repair pixel, and the mapping point occludes the ground point and is occluded by the static point: the mapping point is judged as a suspected dynamic point; After completing the occlusion detection of the current frame elevation image, all mapping points will be judged as one of the following five types: static points, dynamic points, suspected dynamic points, suspected static points, or invalid points; Finally, the occlusion detection results of the elevation images of the first five frames are integrated. If a mapping point is judged to be of the same type more than three times in the five frames, it is finally marked as this type; otherwise, it is marked as a static point.

[0040] In S5 of this implementation, specifically, it includes: The regional consistency check is based on the principle that dynamic point clouds should be far away from static map point clouds. It is used to restore misjudged point clouds. This process performs regional consistency checks on pixels in the first five frames of elevation images that are marked as dynamic points, suspected dynamic points, and suspected static points by occluded detection. During the check, the neighborhood of the current pixel is defined as the 8-pixel area around it. The method of regional consistency check is to determine whether there are points in the neighborhood pixels that meet the conditions. If the Z-axis height difference between a certain type of point in the domain pixel and the marked point is less than the set threshold , then modify the type of the marker point according to specific rules. The principle is as follows Figure 6 As shown, including: S5.1: When the marked point is a dynamic point, if there is a static point that meets the condition in the neighboring pixels, and this condition is met more than three times in the five-frame elevation image, the marked point is re-determined as a static point; otherwise, the marked point is still determined as a dynamic point; S5.2: When the marked point is a suspected dynamic point, if there is a static point that meets the condition in the domain pixel, and this condition is met more than twice in the five-frame elevation image, the marked point is re-determined as a static point; otherwise, the marked point is determined as a dynamic point; S5.3: When the marked point is a suspected static point, if there is a static point that meets the condition in the neighboring pixels, and this condition is met more than once in the five frames of elevation images, the marked point is re-determined as a static point; otherwise, the marked point is determined to be a dynamic point.

[0041] In S6 of this implementation, specifically, it includes: The point cloud that is determined to be a dynamic point after the regional consistency check is clustered using a density-based spatial clustering algorithm with noise (DBSCAN). After clustering is completed, the result is output as a dynamic point cloud.

[0042] Embodiment 2: like Figure 7 As shown, this implementation provides a dynamic point cloud detection system based on elevation images, including: The data acquisition unit is configured to: acquire the ground point cloud, non-ground point cloud and odometer information of each frame of the laser radar point cloud. The specific working method is shown in S1 in Example 1: The elevation image construction unit is configured to: construct an elevation image using the static point cloud and the ground point cloud obtained in the previous frame. The specific working method is shown in S2 in Example 1: The ground continuity repair unit is configured to: perform ground continuity repair on the elevation image. The specific working method is shown in S3 in Example 1: The occlusion detection unit is configured to: perform occlusion detection on the point cloud of the current frame based on the elevation image in combination with the odometer information, and determine dynamic points, suspected dynamic points, and suspected static points. The specific working method is shown in S4 in Example 1: The regional consistency check unit is configured to: perform a regional consistency check on the dynamic point, the suspected dynamic point and the pixel where the suspected static point is located to determine the final dynamic point. The specific working method is shown in S5 in Example 1: The detection result generating unit is configured to: perform a clustering operation on the detected final dynamic point cloud to obtain a dynamic point cloud detection result. The specific working method is shown in S6 in Example 1.

[0043] It is understandable that the above-mentioned units can be separately or completely combined into one or several other units to constitute, or one (some) of the units can be further divided into multiple smaller units in function to constitute, which can achieve the same operation without affecting the realization of the technical effects of the embodiments of the present application. The above-mentioned units are divided based on logical functions. In practical applications, the functions of one unit can also be implemented by multiple units, or the functions of multiple units can be implemented by one unit. In other embodiments of the present application, the system may also include other units. In practical applications, these functions can also be implemented with the assistance of other units, and can be implemented by the collaboration of multiple units.

[0044] According to another embodiment of the present application, the system described in this embodiment can be constructed, and the method of Example 1 of the present application can be implemented by running a computer program (including program code) capable of executing the steps involved in the corresponding method described in Example 1 on a general-purpose computing device such as a computer, which includes processing elements and storage elements such as a central processing unit (CPU), random access memory (RAM), and read-only memory (ROM). The computer program can be recorded on, for example, a computer-readable recording medium, and loaded into the above-mentioned computing device through the computer-readable recording medium and run therein.

[0045] A person skilled in the art can appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed in this application can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. A person skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0046] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on the computer, the process or function according to the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted through a computer-readable storage medium. The computer instructions can be transmitted from a website site, a computer, a server or a data center to another website site, a computer, a server or a data center by wired (e.g., coaxial cable, optical fiber, digital line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that can be accessed by the computer or a data processing device such as a server, a data center, etc. that includes one or more available media integrated. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state drive (SSD)), etc.

[0047] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A dynamic point cloud detection method based on elevation image, characterized in that: The process includes: Get the ground point cloud, non-ground point cloud and odometer information of each frame of lidar point cloud; Construct an elevation image using the static point cloud and ground point cloud obtained in the previous frame; Perform ground continuity repair on elevation images; In combination with the odometer information, occlusion detection is performed on the point cloud of the current frame based on the elevation image to determine dynamic points, suspected dynamic points, and suspected static points; Performing a regional consistency check on the dynamic point, the suspected dynamic point, and the pixels where the suspected static point is located to determine a final dynamic point; The final detected dynamic point cloud is clustered to obtain the dynamic point cloud detection result.

2. The dynamic point cloud detection method based on elevation image according to claim 1, characterized in that: Use the static point cloud and ground point cloud obtained in the previous frame to construct an elevation image, including: Convert each point in the static point cloud and the ground point cloud from a Cartesian coordinate system to a polar coordinate system; According to the resolution of the elevation image, each point is mapped to the corresponding elevation image pixel, and the index of each pixel consists of the ring area number and the sector number; In the elevation image, each pixel records the following information: the number of point clouds contained in the pixel, the coordinate values ​​of the point clouds, and the type of the point clouds; if the pixel contains a ground point cloud, the pixel is marked as a ground pixel; otherwise, it is a non-ground pixel.

3. The dynamic point cloud detection method based on elevation image according to claim 1 or 2, characterized in that: The elevation image includes four annular areas, namely annular area 1, annular area 2, annular area 3 and annular area 4, and the pixel density decreases gradually from the inside to the outside.

4. The dynamic point cloud detection method based on elevation image according to claim 3, characterized in that: Perform ground continuity repair on elevation images, including: The repair process starts from the first ring area, finds all ground pixels in the ring area, and traverses from the ring area number of the pixel from large to small. Starting from each ground pixel, new ground pixels are searched pixel by pixel along the ray direction toward the center of the elevation image. If a new ground pixel is found in annular area 1 or annular area 2, all non-ground pixels between the starting point and the new ground pixel are repaired for ground continuity and marked as ground repair pixels; If no new ground pixels are found in the second annular area, the ray traversal is stopped and the ground continuity repair of the traversed pixels is not performed; After the ground pixel traversal of the annular area 1 is completed, the annular area 2 is processed, all ground pixels in the annular area 2 are extracted, and the pixel annular area numbers are traversed from large to small. Starting from each ground pixel, new ground pixels are searched pixel by pixel along the ray direction toward the center of the elevation image. If a new ground pixel is found in annular area 2 or annular area 3, all non-ground pixels between the starting point and the new ground pixel are repaired for ground continuity and marked as ground repair pixels; If no new ground pixels are found in the third annular area, the ray traversal is stopped and the ground continuity repair of the traversed pixels is not performed; After the ground pixel traversal of the annular area 2 is completed, the annular area 3 is processed, all ground pixels in the annular area 3 are extracted, and the pixel annular area numbers are traversed from large to small. Starting from each ground pixel, new ground pixels are searched pixel by pixel along the ray direction toward the center of the elevation image. If a new ground pixel is found in annular area 3 or annular area 4, ground continuity repair is performed on all non-ground pixels between the starting point and the new ground pixel, and they are marked as ground repair pixels; If no new ground pixel is found in the annular area 4, the ray traversal is stopped and the ground continuity repair of the traversed pixels is not performed; After the traversal of the ground pixels in the annular area three is completed, the annular area four is processed, all the ground pixels in the annular area four are extracted, and they are traversed from large to small according to the pixel ring area number. Starting from each ground pixel, along the direction of the ray toward the center of the elevation image, the ground continuity of all non-ground pixels passed by the ray is repaired, and these pixels are marked as ground repair pixels.

5. The dynamic point cloud detection method based on elevation image according to claim 4, characterized in that: Perform ground continuity repairs, including: According to the distance between the ground repair pixel and the starting ground pixel and the found ground pixel, the ground point cloud height of the ground repair pixel is calculated using the linear interpolation method. The calculation formula is as follows: , where the ground height of the pixel to be repaired is , the ground height of the starting ground pixel is , the ground height of the ground pixel found is , the distance from the pixel to be repaired to the starting ground pixel is , the total distance between the starting ground pixel and the found ground pixel is .

6. The dynamic point cloud detection method based on elevation image according to claim 4, characterized in that: Always maintain the elevation image of the most recent M frames of LiDAR point cloud. In the initialization phase, the time window of the first five frames of LiDAR point cloud is required. During this period, all non-ground point clouds are regarded as static point clouds. Point clouds are only used to construct elevation images without occlusion detection. Starting from the M+1th frame, occlusion detection is performed on the point cloud. Using the odometer data, transform the non-ground point cloud of the current frame into the coordinate system of the previous five frames of elevation images one by one; Dividing the transformed point cloud into corresponding pixels according to the corresponding elevation image resolution, including: first converting the points from the Cartesian coordinate system to the polar coordinate system, and then mapping each point to the corresponding pixel according to the resolution of the elevation image; Perform occlusion detection on the pixel where the point to be detected is located in the elevation image. If the difference between the Z-axis height of the point to be detected and the Z-axis height of the highest ground point in the pixel exceeds the set threshold , then it is considered that the point to be detected has blocked the ground point; otherwise, it is considered that there is no occlusion; if the difference between the Z-axis height of the static point in the pixel and the Z-axis height of the point to be detected exceeds the threshold , then the point to be detected is considered to be blocked by the static point; otherwise, it is considered that no blockage occurs.

7. The dynamic point cloud detection method based on elevation image according to claim 6, characterized in that: During the occlusion detection process, the following situations occur: The pixel is a non-ground pixel and there is no point cloud in the pixel: the mapping point is determined to be a suspected static point; The pixel is a non-ground pixel and the mapping point occludes the static object in the pixel: the mapping point is judged as a static point; The pixel is a non-ground pixel and the mapping point is occluded by a static object in the pixel: the mapping point is determined to be a static point; The pixel is a ground pixel or a ground repair pixel and the mapping point is blocked by the ground point in the pixel: the mapping point is judged as an invalid point; The pixel is a ground pixel or a ground repair pixel and the mapping point only occludes the ground point: the mapping point is determined to be a dynamic point; The pixel is a ground pixel or a ground repair pixel and the mapping point occludes the ground point and is occluded by the static point: the mapping point is determined to be a suspected dynamic point; After completing the occlusion detection of the current frame elevation image, all mapping points are judged as one of the following five types: static points, dynamic points, suspected dynamic points, suspected static points, or invalid points; The occlusion detection results of the elevation images of the first M frames are integrated. If a mapping point is judged to be of the same type more than (M+1) / 2 times in five frames, it is finally marked as this type; otherwise, it is marked as a static point.

8. The dynamic point cloud detection method based on elevation image according to claim 6, characterized in that: Performing a regional consistency check on the dynamic point, the suspected dynamic point, and the pixel where the suspected static point is located to determine the final dynamic point includes: The neighborhood of the current pixel is defined as the 8 pixel area around it, and whether there are points that meet the conditions in the neighborhood pixels. If the Z-axis height difference between a certain type of point in the domain pixel and the marked point is less than the set threshold , the type of the marker point is modified according to specific rules, including: When the marked point is a dynamic point, if there is a static point that meets the conditions in the neighboring pixels, and this condition is met more than three times in the five-frame elevation image, the marked point will be re-determined as a static point; otherwise, the marked point is still determined as a dynamic point; When the marked point is a suspected dynamic point, if there is a static point that meets the conditions in the domain pixels, and this condition is met more than twice in the five-frame elevation image, the marked point is re-determined as a static point; otherwise, the marked point is determined as a dynamic point; When a marked point is a suspected static point, if there is a static point that meets the conditions in the neighborhood pixels and this condition is met more than once in the five-frame elevation image, the marked point is re-determined as a static point; otherwise, the marked point is determined to be a dynamic point.

9. The dynamic point cloud detection method based on elevation image according to claim 1 or 2, characterized in that: For the point clouds that are determined to be dynamic points after the regional consistency check, a density-based spatial clustering algorithm with noise is used to perform clustering operations.

10. A dynamic point cloud detection system based on elevation image, characterized in that: include: The data acquisition unit is configured to: acquire ground point cloud, non-ground point cloud and odometer information of each frame of laser radar point cloud; The elevation image construction unit is configured to: construct an elevation image using a static point cloud and a ground point cloud obtained in a previous frame; The ground continuity repair unit is configured to: perform ground continuity repair on the elevation image; An occlusion detection unit is configured to: perform occlusion detection on the point cloud of the current frame based on the elevation image in combination with the odometer information, and determine dynamic points, suspected dynamic points, and suspected static points; A regional consistency check unit is configured to: perform a regional consistency check on the dynamic point, the suspected dynamic point and the pixel where the suspected static point is located to determine a final dynamic point; The detection result generating unit is configured to: perform a clustering operation on the detected final dynamic point cloud to obtain a dynamic point cloud detection result.

Images