Static point cloud map construction method based on density scan ratio test and neighborhood inspection

Abstract

The invention discloses a static point cloud map construction method based on density scan ratio test and neighborhood inspection. Firstly, a query frame and a global point cloud map are processed according to straight-through filtering, and a region of interest is obtained; secondly, performing grid division, and performing density scanning ratio test according to the same sub-regions of the query frame and the global point cloud map to extract a suspected dynamic region; querying a dynamic sub-region and a static sub-region around a neighborhood of the suspected dynamic sub-region around the suspected dynamic sub-region, judging an occupation attribute around the suspected dynamic sub-region, converting the suspected dynamic sub-region into the determined dynamic sub-region and the determined static sub-region, converting the static sub-region around the dynamic sub-region into the determined static sub-region and the determined dynamic sub-region; and finally, removing dynamic point clouds in the dynamic sub-regions, and constructing a global static point cloud map. According to the method, the dynamic information in the environment can be accurately extracted and eliminated, the static point cloud can be prevented from being mistakenly removed, and the construction precision of the static point cloud map can be improved.

Description

Technical Field

[0001] The present invention belongs to the field of navigation and positioning for autonomous driving, and particularly relates to a method for constructing a static point cloud map based on density scan ratio test and neighborhood check. Background Art

[0002] A stable and reliable point cloud map is crucial for an autonomous driving vehicle to perform high-precision navigation and path planning using multi-source sensors. Compared with traditional two-dimensional maps, three-dimensional point cloud maps contain rich depth information and details, such as road surface undulations, obstacle heights, and road boundaries, thus greatly improving the accuracy of environmental modeling. Especially in areas where satellite signals are weak or lost, the vehicle can achieve centimeter-level high-precision positioning through data matching between the point cloud map and lidar sensors, ensuring safe driving. In addition, the point cloud map provides reliable support for path planning. By providing accurate road structure, lane width, and slope change information, it helps the vehicle quickly adjust its driving route when dealing with dynamic road conditions, ensuring driving efficiency and safety. The point cloud map also provides key support for multi-source sensor fusion. By combining data from sensors such as lidar, cameras, and radars, it enables a more comprehensive environmental perception. Even in adverse weather or lighting conditions, the lidar can still work normally, and after fusing with camera information, the overall navigation reliability is improved.

[0003] In some scenarios, as the point cloud map is generated, the trajectories of dynamic objects frequently appear in the map, forming "ghosts" on the map. These "ghosts" on the map can bring multiple negative impacts to the autonomous driving system, significantly reducing the reliability and navigation accuracy of the map. First, the residual trajectories of these dynamic objects, usually due to pedestrians, vehicles, or other moving objects being captured and stored during map generation, will cause the point cloud map to contain false information that does not belong to the real environment, resulting in misjudgment by the autonomous driving vehicle during path planning and obstacle avoidance. It may misidentify the ghost as an actual obstacle, thus triggering unnecessary avoidance behaviors and affecting the vehicle's driving efficiency. In addition, these false object trajectories will also trouble the environmental perception module, increasing the difficulty of sensor data fusion and processing, making it more difficult for the autonomous driving system to accurately distinguish real static obstacles and dynamic threats, and thus threatening driving safety. Therefore, it is necessary to study a method for generating a static point cloud map that can remove dynamic point clouds while preventing the accidental deletion of static point clouds. Summary of the Invention

[0004] To solve the above problems, the present invention proposes a method for constructing a static point cloud map based on density scan ratio test and neighborhood check, which can effectively remove dynamic point clouds in the environment and prevent the accidental deletion of static point clouds, and construct a clear and complete static point cloud map.

[0005] To achieve the above object, the technical solution of the present invention is as follows: In a first aspect, the present invention provides a static point cloud map construction method based on density scanning ratio test and neighborhood check, comprising the following steps:

[0006] S1: Extracting the region of interest of the query frame and the global point cloud map based on the through filtering;

[0007] S2: Divide the region of interest into grids;

[0008] S3: Perform density-assisted scan ratio test on the same sub-regions of the frame and map to extract suspected dynamic sub-regions;

[0009] S4: Check the neighborhood of the suspected dynamic sub-region to see if there are other suspected dynamic sub-regions around it. If so, convert the suspected dynamic sub-region into a real dynamic sub-region. If not, convert it into a static region. Check the static sub-regions around the suspected dynamic sub-region. If the static sub-region is in the middle of the suspected dynamic sub-region, it is also a dynamic region.

[0010] S5: Perform ground segmentation on the detected dynamic sub-area, extract the non-ground point cloud, and use it as the dynamic point cloud of the sub-area. Finally, remove the dynamic point cloud detected on the global point cloud map and construct a static point cloud map.

[0011] In a second aspect, the present invention provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the method described in the first aspect when executing the program.

[0012] In a third aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in the first aspect.

[0013] In a fourth aspect, the present invention provides a computer program product, comprising a computer program, which implements the steps of the method described in the first aspect when executed by a processor.

[0014] Compared with the prior art, the present invention has the following beneficial effects:

[0015] First of all, the present invention takes into account that dynamic objects are usually not as far away as possible, so a method based on fixed grids is used to divide the region of interest. Compared with the traditional grid division scheme, the present invention is more capable of preventing static points from being deleted by mistake. In addition, compared with the traditional scanning ratio test, the present invention combines density ratio assistance on this basis, which can more accurately detect suspected dynamic areas, rather than treating many occluded areas as dynamic areas. This method alleviates the deficiency of scanning ratio test's sensitivity to occlusion to a certain extent and improves the overall operating efficiency.

[0016] Secondly, the present invention uses a neighborhood detection method to re-determine the suspected dynamic region. Neighborhood dynamic checking can prevent static sub-regions from being misdeleted as dynamic sub-regions, which can effectively maintain the integrity of the static point cloud map. Neighborhood static checking effectively prevents the situation where parts of dynamic objects are misjudged as static point clouds due to perspective problems, and can effectively detect static point clouds according to the consistency of the space occupied by dynamic objects, improving the accuracy of dynamic point cloud detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a block diagram of the method for constructing a static point cloud map based on density scan ratio test and neighborhood check of the present invention.

[0018] Figure 2 It is the result of the static mapping comparison experiment of the present invention on the SemanticKITTI dataset.

[0019] Figure 3 and Figure 4 It is the comparison result before and after removing dynamic objects from the point cloud in the actual measurement environment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0020] The present invention will be further clarified below in conjunction with the drawings and specific embodiments. It should be understood that the following specific embodiments are only used to illustrate the present invention and not to limit the scope of the present invention.

[0021] A method for constructing a static point cloud map based on density scan ratio test and neighborhood check according to the present invention has the implementation principle as Figure 1 shown, and its process mainly includes the following steps:

[0022] Step S1, extract the region of interest of the query frame and the global point cloud map according to the pass-through filter;

[0023] Extract the region of interest from the original query frame and the global point cloud map. Taking the query frame as an example, for any point in the t-th moment If the distances from the three axes of the lidar satisfy a certain relationship, then is a point in the region of interest, and the region of interest can be expressed as:

[0024]

[0025] Here, L max represents the maximum horizontal range of the region of interest, is the distance of the point from the lidar in the x and y directions, and H minand H max represent the minimum and maximum ranges of interest on the z-axis.

[0026] Step S2, divide the region of interest into grids;

[0027] Process S1 obtained the horizontal range L max ×L max . According to the fixed length L, divide it into N x ×N y . To prevent negative indexing, we move the point cloud center to (-L max , -L max ). A t,(i,j) represents the grid square in the (i, j)-th region of interest at time t and can be expressed as:

[0028]

[0029] Finally, the grid squares of interest in the query frame can be expressed as The grid squares of interest in the global point cloud map can be expressed as

[0030] Step S3, perform a density-assisted scan ratio test on the same sub-regions of the frame and the map, and extract suspected dynamic sub-regions;

[0031] Specifically, it includes the following process:

[0032] Taking the (i, j)-th sub-region as an example, check the height difference ratio of the same query frame and the same region of the map, which can be expressed as ΔH:

[0033]

[0034] Here, and represent the maximum and minimum height differences of the (i, j)-th sub-region of the query frame and the global map.

[0035] Then calculate the density ratio ΔN of the (i, j)-th sub-region, which can be expressed as:

[0036]

[0037] Among them, represents the density ratio of the sub-region at the i-th row and j-th column of the map grid at time t, represents the density ratio of the sub-region at the i-th row and j-th column of the query frame grid at time t. Specifically, it can be expressed as:

[0038]

[0039]

[0040] Here, N(·) represents the number of recorded conditions, represents the z - value of a certain point in the sub - region of the i - th row and j - th column of the map grid at time t, represents the z - value of a certain point in the sub - region of the i - th row and j - th column of the query - frame grid at time t, and h is the fixed height threshold of the occupied object, which is generally selected as 1m according to engineering experience.

[0041] Finally, we can obtain the density - assisted scan - ratio test D - SRT t,(i,j) The expression of which is:

[0042] D - SRT t,(i,j) = ΔHe 1-ΔN

[0043] Through the density - assisted scan - ratio test, a set of suspected dynamic sub - regions can be obtained:

[0044]

[0045] Here, represents the dynamic detection threshold of the density - assisted scan - ratio test, which is usually set by experience and taken as 0.2.

[0046] Step S4: Conduct a neighborhood check on the suspected dynamic sub - regions to check whether there are other suspected dynamic sub - regions around them. If there are, convert the suspected dynamic sub - regions into real dynamic sub - regions; if not, convert them into static regions. Check the static sub - regions around the suspected dynamic sub - regions. If a static sub - region is in the middle of a suspected dynamic sub - region, it is also a dynamic region.

[0047] Based on the continuity of the occupied space, conduct a neighborhood check on the suspected dynamic sub - regions to check whether there are other suspected dynamic sub - regions around them. If there are, convert the suspected dynamic sub - regions into real dynamic sub - regions; if not, convert them into static regions. And check the static sub - regions around the suspected dynamic sub - regions. If a static sub - region is in the middle of a suspected dynamic sub - region, it is also a dynamic region.

[0048] For the suspected dynamic regions If the adjacent 8 - neighborhood is all static, then it is an isolated dynamic point cloud. If there are dynamic sub - regions in the adjacent 8 - neighborhood, then it is a dynamic sub - region, otherwise it is static. It can be expressed as:

[0049]

[0050] Here, "dyn" represents that the sub - region is dynamic, and "sta" represents that the sub - region is static. represents taking The n-neighborhood set centered on f(·) = sus represents the suspected dynamic point cloud.

[0051] For The static sub-region around the area If the x-axis or y-axis neighborhood of this static sub-region is a suspected dynamic region, then this region is also dynamic, otherwise it is static. It can be expressed as:

[0052]

[0053] where || represents the OR logical operator.

[0054] Step S5, perform ground segmentation on the detected dynamic sub-region, extract the non-ground point cloud, use it as the dynamic point cloud of this sub-region, and finally remove the dynamic point cloud detected on the global point cloud map to construct a static point cloud map.

[0055] For the detected dynamic region, extract the ground seed points for pre-ground point extraction, and then use the method of plane fitting to extract more accurate ground points.

[0056] Figure 2 This is the result of the static mapping comparison experiment of the present invention on the SemanticKITTI dataset. The first three, Octomap, Removert, and ERASOR, are advanced methods for constructing static point cloud maps. It can be seen that our method removes more dynamic point clouds (green trajectories) compared to the first three methods and retains more static point clouds. In the figure, the red color represents the situation where static point clouds are wrongly removed. The proposed DNC-REMOVER algorithm of ours retains more correct static point clouds, which reflects that the proposed algorithm can correctly remove dynamic points and retain more static point clouds.

[0057] Figure 3 It represents the schematic diagram of the actual vehicle in the campus vehicle experiment. The scene contains the movement trajectories caused by vehicles, electric vehicles, and pedestrians, aiming to verify the effectiveness of the proposed algorithm through a complex dynamic environment. Figure 4 In (a), it represents the point cloud map generated by LIOSAM in this scene. To show it more clearly, we removed the ground to display the dynamic trajectories of vehicles and pedestrians. (b) represents the situation after vehicles and pedestrians are successfully removed. It can be seen that the dynamic trajectories within the red frame are significantly reduced.

[0058] It should be noted that the above content only illustrates the technical idea of the present invention and cannot be used to limit the protection scope of the present invention. For those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements can still be made, and these improvements and refinements all fall within the protection scope of the claims of the present invention.

Claims

1. A static point cloud map construction method based on density scanning ratio test and neighborhood check, characterized in that: The following steps are involved: S1: Extracting the region of interest of the query frame and the global point cloud map based on the through filtering; S2: Divide the region of interest into grids; S3: Perform density-assisted scan ratio test on the same sub-regions of the frame and map to extract suspected dynamic sub-regions; S4: Check the neighborhood of the suspected dynamic sub-region to see if there are other suspected dynamic sub-regions around it. If so, convert the suspected dynamic sub-region into a real dynamic sub-region. If not, convert it into a static region. Check the static sub-regions around the suspected dynamic sub-region. If the static sub-region is in the middle of the suspected dynamic sub-region, it is also a dynamic region. S5: Perform ground segmentation on the detected dynamic sub-area, extract the non-ground point cloud, and use it as the dynamic point cloud of the sub-area. Finally, remove the dynamic point cloud detected on the global point cloud map and construct a static point cloud map.

2. The static point cloud map construction method based on density scanning ratio test and neighborhood check according to claim 1 is characterized in that: The step S1 specifically The process includes: Extract the region of interest from the original query frame and the global point cloud map; for the query frame t time Any point in if The three-axis distance from the laser radar satisfies a certain relationship, then is a point in the region of interest, and the region of interest is expressed as: Here, L max represents the maximum horizontal extent of the region of interest, For point The distance from the laser radar in the x and y directions, H min and H max Indicates the minimum and maximum range of interest on the z-axis.

3. The static point cloud map construction method based on density scanning ratio test and neighborhood check according to claim 1 is characterized in that: The step S2 specifically includes the following process: Step S1 obtains the horizontal range L of the region of interest max ×L max ; According to the fixed length L, it is divided into N x ×N y ; To prevent negative indexing, the point cloud center is moved to (-L max ,-L max );A t,(i,j) Represents the square in the (i, j)th region of interest at time t, which can be expressed as: Finally, the interesting square of the query frame is expressed as The grid of interest in the global point cloud map is represented as 4. The static point cloud map construction method based on density scanning ratio test and neighborhood check according to claim 2 is characterized in that: The step S3 specifically The process includes: For the (i, j)th sub-region, check the height difference ratio between the same query frame and the same area of ​​the map, expressed as ΔH: in, and Indicates the maximum and minimum height differences between the query frame and the (i, j)th sub-region of the global map; Then the density ratio ΔN of the (i, j)th sub-region is calculated, which is expressed as: in, represents the density ratio of the sub-area in the i-th row and j-th column of the map grid at time t, It represents the density ratio of the sub-region of the i-th row and j-th column of the query frame grid at time t; it is specifically expressed as: Where N(·) represents the number of recorded conditions, It represents the z value of a point in the sub-area of ​​the i-th row and j-th column of the map grid at time t. It means to query the z value of a point in the sub-region of the i-th row and j-th column of the frame grid at time t, where h is the fixed height threshold of the occupied object; Get Density Assisted Scanning Ratio Test D-SRT t,(i,j) The expression is: D-SRT t,(i,j) =ΔHe 1-ΔN Through the density-assisted scanning ratio test, a set of suspected dynamic sub-regions can be obtained: here, Indicates the dynamic detection threshold for density-assisted scan ratio testing.

5. The static point cloud map construction method based on density scanning ratio test and neighborhood check according to claim 4 is characterized in that: Take 0.

2.

6. The static point cloud map construction method based on density scanning ratio test and neighborhood check according to claim 1 is characterized in that: The step S4 specifically includes the following process: Based on the continuity of the occupied space, the suspected dynamic sub-region is checked for its neighborhood to see if there are other suspected dynamic sub-regions around it. If so, the suspected dynamic sub-region is converted into a real dynamic sub-region. If not, it is converted into a static region. The static sub-regions around the suspected dynamic sub-region are checked. If the static sub-region is in the middle of the suspected dynamic sub-region, it is also a dynamic region. For suspected dynamic areas If the adjacent 8-neighborhoods are all static, then is an isolated dynamic point cloud; if there is a dynamic sub-region in the adjacent 8 neighborhoods, then is a dynamic sub-region, otherwise it is static; it is expressed as: Here "dyn" means the sub-region is dynamic, and "sta" means the sub-region is static; Indicates is a set of n neighborhoods centered on the cloud, f(·)=sus represents a suspected dynamic point cloud; for Static sub-regions around the region If the x-axis or y-axis neighbors of the static sub-region are all suspected dynamic regions, then the region is also dynamic, otherwise it is static; it is expressed as: Among them, || represents the OR logical operator.

7. A computer device comprising a memory, a processor and a computer program in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps of the method according to any one of claims 1 to 6 are implemented.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method described in any one of claims 1 to 6 are implemented.

9. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method described in any one of claims 1 to 6 are implemented.

Images