CN-122023166-A - Point cloud noise filtering method and device and related equipment

CN122023166ACN 122023166 ACN122023166 ACN 122023166ACN-122023166-A

Abstract

The application discloses a point cloud noise filtering method, a device and related equipment, wherein the method comprises the steps of downsampling first point cloud data to obtain second point cloud data, wherein the first point cloud data is obtained by detecting a target area by a laser radar; the method comprises the steps of selecting a candidate noise point cloud set from the second point cloud data, wherein the distance between each point cloud in the candidate noise point cloud set is within a preset distance range, the reflection intensity of each point cloud is within a preset first intensity range, judging whether each point cloud in each voxel is noise or not based on the number of the point clouds in each voxel of a point cloud grid, the positions of each point cloud and the reflection intensity, wherein the point cloud grid is generated based on each point cloud in the candidate noise point cloud set, and removing the point cloud judged to be noise from the candidate noise point cloud set to obtain target point cloud data. The application can effectively filter eVTOL the received miscellaneous points in the vertical take-off and landing and low-altitude flight, and improves the stability of obstacle identification.

Inventors

SU QINGPENG
ZHOU XIN

Assignees

广东高域科技有限公司

Dates

Publication Date: 20260512
Application Date: 20251218
Priority Date: 20250829

Claims (10)

1.A method for filtering point cloud noise, comprising: Screening candidate noise point clouds from the second point cloud data, wherein the distance between each point cloud in the candidate noise point clouds is within a preset distance range, and the reflection intensity of each point cloud is within a preset first intensity range; Judging whether each point cloud in each voxel is noise or not based on the number of the point clouds, the positions and the reflection intensity of the point clouds in each voxel of a point cloud grid, wherein the point cloud grid is generated based on the point clouds in the candidate noise point cloud set; and eliminating the point cloud judged as noise from the candidate noise point cloud set to obtain target point cloud data.
2. The method according to claim 1, wherein the method further comprises: And performing downsampling processing on the first point cloud data by using a voxel filtering method based on a preset first voxel size to obtain the second point cloud data, wherein the first point cloud data is obtained by detecting a target area by a laser radar.
3. The method of claim 1, wherein the process of screening candidate noise point clouds from the second point cloud data comprises: traversing the point cloud in the second point cloud data, and for the currently traversed point cloud: Judging whether the distance of the point cloud is not larger than a preset distance threshold value; if yes, judging whether the reflection intensity of the point cloud is smaller than a preset first intensity threshold value; if yes, determining the point cloud as a candidate noise point cloud; and forming a candidate noise point cloud set by each candidate noise point cloud.
4. The method of claim 1, wherein generating a point cloud grid based on each point cloud in the set of candidate noise point clouds comprises: Generating a voxel grid based on a preset second voxel size and the space where each point cloud in the candidate noise point cloud set is located; Determining voxels to which each point cloud belongs based on the position of each point cloud in the candidate noise point cloud set; And forming a point cloud grid by the voxel grid and the point cloud contained in each voxel in the voxel grid.
5. The method of claim 1, wherein determining whether each point cloud in a voxel is noisy based on a number of point clouds in each voxel, a location of each point cloud, and a reflection intensity, comprises: traversing voxels in the point cloud grid, for the currently traversed voxels: acquiring the number of point clouds in the voxels, and the position and the reflection intensity of each point cloud; Acquiring the position variance of each point cloud in the voxel based on the position of each point cloud in the voxel; Acquiring the average reflection intensity and the reflection intensity variance of each point cloud in the voxel based on the reflection intensity of each point cloud in the voxel; And judging whether the point clouds in the voxels are noise or not based on the number of the point clouds of the voxels, the position variance, the average reflection intensity and the reflection intensity variance of the point clouds in the voxels.
6. The method of claim 5, wherein determining whether the point cloud in the voxel is noisy based on the number of point clouds of the voxel, a location variance of the point clouds in the voxel, an average reflection intensity, and a reflection intensity variance, comprises: Judging whether the number of the point clouds of the voxels is larger than a preset number threshold; If yes, judging whether the position variance of each point cloud in the voxels is larger than a preset position divergence threshold value; if yes, judging whether the average reflection intensity of each point cloud in the voxels is smaller than a preset second intensity threshold value; If yes, judging whether the reflection intensity variance of each point cloud in the voxels is smaller than a preset intensity divergence threshold value; if yes, determining that each point cloud in the voxels is noise.
7. The method according to claim 5 or 6, wherein the process of obtaining the position variance of each point cloud in the voxel based on the position of each point cloud in the voxel comprises: Acquiring three-dimensional coordinates of each point cloud in the voxels; Acquiring an average center coordinate of the voxel based on the three-dimensional coordinate of each point cloud in the voxel; acquiring the variance of each point cloud in the voxel on the basis of the average center coordinates in the voxel and the three-dimensional coordinates of the point cloud; and acquiring the position variance of each point cloud in the voxel based on the variance of each dimension in the voxel.
8. A point cloud noise filtering apparatus, comprising: The rough filtering unit is used for screening candidate noise point clouds from the second point cloud data, the distance between each point cloud in the candidate noise point clouds is within a preset distance range, and the reflection intensity of each point cloud is within a preset first intensity range; the noise identification unit is used for judging whether each point cloud in each voxel is noise or not based on the number of the point clouds, the positions of the point clouds and the reflection intensity in each voxel of the point cloud grid, and the point cloud grid is generated based on the point clouds in the candidate noise point cloud set; And the fine filtering unit is used for removing the point cloud judged to be noise from the candidate noise point cloud set to obtain target point cloud data.
9. The point cloud noise filtering device is characterized by comprising a memory and a processor; The memory is used for storing programs; the processor is configured to execute the program to implement the steps of the point cloud noise filtering method according to any one of claims 1 to 7.
10. A storage medium having stored thereon a computer program which, when executed by a processor, implements the steps of the point cloud noise filtering method according to any of claims 1 to 7.

Description

Point cloud noise filtering method and device and related equipment Technical Field The application relates to the technical field of low-altitude flight, in particular to a point cloud noise filtering method, a point cloud noise filtering device and related equipment. Background EVTOL (ELECTRIC VERTICAL TAKE-off AND LANDING, electric vertical take-off and landing aircraft) is a novel vehicle capable of taking off and landing vertically without a runway, has the characteristics of zero emission, low noise and the like, and is regarded as an important carrier for low-altitude economy. eVTOL in low altitude flight, it is desirable to have aerial small target recognition capability to improve the safety of the aircraft, such as recognition of small unmanned aerial vehicles, birds, etc., where high resolution lidar is required. However, high resolution lidars are particularly sensitive to small objects, for which extraneous small objects can contribute a lot of noise. For example, when performing vertical take-off and landing with multiple rotors eVTOL, the airflow generated by the rotation of the blades may wind up dust or grass clippings on the ground, and for example, during a eVTOL low-altitude flight, it is also possible to wind up nearby insect flocks (such as water ant flocks) into the airflow. The dust, grass clippings, insect flocks, etc. can create a lot of noise point clouds near eVTOL aircraft, thereby affecting obstacle detection and reducing signal to noise ratio. Therefore, how to filter noise of irrelevant small objects becomes a technical problem to be solved. Disclosure of Invention In view of the above, the application provides a method, a device and related equipment for filtering point cloud noise, so as to filter eVTOL noise miscellaneous points received during vertical take-off, landing and low-altitude flight. To achieve the above object, a first aspect of the present application provides a method for filtering point cloud noise, including: Screening candidate noise point clouds from the second point cloud data, wherein the distance between each point cloud in the candidate noise point clouds is within a preset distance range, and the reflection intensity of each point cloud is within a preset first intensity range; Judging whether each point cloud in each voxel is noise or not based on the number of the point clouds, the positions and the reflection intensity of the point clouds in each voxel of a point cloud grid, wherein the point cloud grid is generated based on the point clouds in the candidate noise point cloud set; and eliminating the point cloud judged as noise from the candidate noise point cloud set to obtain target point cloud data. Preferably, the method further comprises: and performing downsampling processing on the first point cloud data by using a voxel filtering method based on a preset first voxel size to obtain second point cloud data, wherein the first point cloud data is obtained by detecting a target area by a laser radar. Preferably, the process of screening candidate noise point clouds from the second point cloud data includes: traversing the point cloud in the second point cloud data, and for the currently traversed point cloud: Judging whether the distance of the point cloud is not larger than a preset distance threshold value; if yes, judging whether the reflection intensity of the point cloud is smaller than a preset first intensity threshold value; if yes, determining the point cloud as a candidate noise point cloud; and forming a candidate noise point cloud set by each candidate noise point cloud. Preferably, the process of generating the point cloud grid based on the point clouds in the candidate noise point cloud set includes: Generating a voxel grid based on a preset second voxel size and the space where each point cloud in the candidate noise point cloud set is located; Determining voxels to which each point cloud belongs based on the position of each point cloud in the candidate noise point cloud set; And forming a point cloud grid by the voxel grid and the point cloud contained in each voxel in the voxel grid. Preferably, the process of judging whether each point cloud in each voxel is noise based on the number of the point clouds, the positions of the point clouds and the reflection intensity in each voxel of the point cloud grid comprises the following steps: traversing voxels in the point cloud grid, for the currently traversed voxels: acquiring the number of point clouds in the voxels, and the position and the reflection intensity of each point cloud; Acquiring the position variance of each point cloud in the voxel based on the position of each point cloud in the voxel; Acquiring the average reflection intensity and the reflection intensity variance of each point cloud in the voxel based on the reflection intensity of each point cloud in the voxel; And judging whether the point clouds in the voxels are noise or not based on the number of the point