Point cloud denoising method
By dividing the point cloud into sub-neighborhoods and performing bilateral filtering, combined with radius filtering, the impact of step edge on point cloud noise reduction in the existing technology is solved, and edge preservation performance and accuracy are improved, which is suitable for aircraft component detection. , improves detection accuracy and efficiency.
Patent Information
- Application Number
- CN202310748861.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-06-25
- Publication Date
- 2024-11-19
- Estimated Expiration
- 2043-06-25
AI Technical Summary
The existing point cloud noise reduction method is difficult to ensure edge preservation performance and noise reduction accuracy when processing step edges, which affects the accuracy of digital measurement of structured light and the quality of aircraft parts detection.
By dividing the neighborhood point clouds of the point cloud to be filtered into multiple sub-neighborhoods and performing bilateral filtering, combining radius filtering, the low-frequency and high-frequency areas are directly separated at the step edge for filtering, ensuring edge preservation and accuracy.
It significantly improves the edge preservation performance and accuracy of point cloud noise reduction, reduces processing complexity, improves detection accuracy and efficiency, and is suitable for large-scale component measurements in aircraft manufacturing.
Smart Images

Figure CN116843563B_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of digital measurement, and in particular relates to a point cloud noise reduction processing method. Background Art
[0002] As the aviation manufacturing industry continues to develop in the direction of intelligence and digitalization, the quality inspection requirements for aviation parts are also constantly increasing. The degree of fit between the manufacturing dimensions and the design dimensions of aircraft parts directly reflects the quality of aircraft products. Therefore, high-precision inspection of the dimensions of aircraft parts will play a vital role in improving the quality of aircraft products. Traditional aircraft parts inspection methods still rely on conventional tools such as vernier calipers and feeler gauges to measure the dimensions, which cannot accurately detect the dimensions of parts, and it is also difficult to form large-scale measurement data to guide the optimization of parts processing.
[0003] With the development of 3D measurement technology, digital detection technologies such as photogrammetry, time-of-flight measurement, and structured light measurement have been gradually applied to the inspection of aircraft parts. Among them, structured light measurement technology mainly obtains point cloud data on the surface of parts through 3D scanning equipment, and processes the point cloud data to improve measurement accuracy. In structured light measurement technology, point cloud noise reduction is the most critical link in point cloud data processing. The processing accuracy of point cloud noise reduction will directly affect the accuracy of structured light digital measurement, and thus affect the product quality of the aircraft.
[0004] Patent document CN113177897A discloses a fast lossless filtering method for disordered 3D point clouds, which performs gridding on the circumscribed cube of the point cloud and converts the dense and disordered point cloud into a regular binary three-dimensional point cloud, thereby reducing the amount of calculation. In addition, patent document CN103853840A discloses a method for filtering uneven scattered point cloud data, which indexes and organizes the scattered point cloud data, organizes the scattered point clouds into a regular cube grid model, and obtains the scattered measurement points contained in each grid, thereby filtering the scattered points. However, the existing point cloud denoising methods do not consider the impact of step edges on the point cloud denoising accuracy, which affects the edge preservation performance of the point cloud denoising, and the denoising accuracy is difficult to guarantee. Summary of the invention
[0005] The purpose of the present invention is to provide a point cloud denoising processing method, which can solve the influence of step edges on point cloud denoising, improve the edge preservation of point cloud denoising at step edges, and improve the accuracy of point cloud denoising.
[0006] The present invention is achieved through the following technical solutions:
[0007] The point cloud denoising processing method includes the following steps:
[0008] A predefined bounding cube is used to divide the neighborhood point cloud into multiple sub-neighborhood point clouds with the point cloud to be filtered as the center;
[0009] Perform bilateral filtering on each sub-neighborhood point cloud data to obtain a new point cloud;
[0010] For the filtered point cloud and the new point cloud, radius filtering is performed with the filtered point cloud as the center to complete the denoising of the point cloud.
[0011] In some embodiments, the method for obtaining the neighborhood point cloud data of the point cloud to be filtered is: obtaining the original point cloud data, and using the K nearest neighbor algorithm to obtain the neighborhood point cloud data of the point cloud to be filtered.
[0012] In some embodiments, a structured light system is used to obtain point cloud data of the object to be processed at different viewing angles, and the point cloud data at different viewing angles are stitched together using a point cloud stitching algorithm to obtain original point cloud data.
[0013] In some embodiments, the method of dividing a neighborhood point cloud into a plurality of sub-neighborhood point clouds is: defining eight bounding box cubes, placing the eight bounding box cubes together, and making their common intersections the point clouds to be filtered.
[0014] In some embodiments, the side length of each bounding cube is greater than the radius of the neighborhood point cloud.
[0015] In some embodiments, the method of using bilateral filtering to filter each sub-neighborhood point cloud data to obtain a new point cloud is:
[0016] Suppose the point cloud to be filtered is p, the neighborhood point cloud set of p is N(p), and each sub-neighborhood point cloud is N i (p);
[0017] Then the normal vector of bilateral filtering at point cloud p is:
[0018]
[0019] in, is the unit normal vector of point cloud p, and its position is represented by l p , is the unit normal vector of the neighborhood point cloud of point cloud p; W C is a standard Gaussian filter, W S is a standard Gaussian filter,
[0020] is the unit normal vector and The intensity difference between them is expressed as:
[0021] After bilateral filtering of each sub-neighborhood point cloud set, the new point cloud data obtained can be expressed as:
[0022]
[0023] in, Represents the sub-neighborhood point cloud location.
[0024] In some embodiments, when radius filtering is performed on the to-be-filtered point cloud and the new point cloud, when the point cloud data in the circle is equal to 1, the point cloud is discarded.
[0025] In some embodiments, when radius filtering is performed on the point cloud to be filtered and the new point cloud, when the number of point clouds in the circle is greater than 1, bilateral filtering is performed again on the point cloud data in the circle.
[0026] Compared with the prior art, the present invention has the following advantages and beneficial effects:
[0027] The present invention divides the neighborhood points of the point cloud to be filtered into different sub-neighborhoods and performs filtering processing separately, which can greatly improve the edge preservation performance of filtering and the accuracy of filtering processing. It has good practicality and realistic significance in the application of dimensional measurement of a large number of products in aircraft manufacturing, and can greatly improve the processing speed in the detection process, improve the detection accuracy, and at the same time improve the detection efficiency, thereby ensuring the delivery cycle of parts.
[0028] The present invention directly divides the neighborhood of each point cloud into different areas to select low-frequency areas and high-frequency areas at the step edge, and performs bilateral filtering on them respectively, thereby reducing the complexity of the processing method and improving its robustness, and can well realize edge-preserving filtering. BRIEF DESCRIPTION OF THE DRAWINGS
[0029] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.
[0030] Figure 1 This is a flow chart of the point cloud denoising processing method of the present invention.
[0031] Figure 2 This is a schematic diagram of the present invention using a border cube to process a neighborhood point cloud to be filtered.
[0032] Figure 3a ) is a schematic diagram of a case where the present invention performs radius filtering on the filtered point cloud and the new point cloud.
[0033] Figure 3b ) is a schematic diagram of another case in which the present invention performs radius filtering on the filtered point cloud and the new point cloud. DETAILED DESCRIPTION
[0034] In order to make the purpose, technical scheme and advantages of the present application clearer, the specific embodiments of the present application are further described in detail below in conjunction with the accompanying drawings. It is understood that the specific embodiments described herein are only used to explain the present application, rather than to limit the present application. It should also be noted that, for the convenience of description, only the part related to the present application but not all the contents are shown in the accompanying drawings. Before discussing the exemplary embodiments in more detail, it should be mentioned that some exemplary embodiments are described as processes or methods depicted as flow charts. Although the flow chart describes each operation (or step) as a sequential process, many of the operations therein can be implemented in parallel, concurrently or simultaneously. In addition, the order of each operation can be rearranged. The process can be terminated when its operation is completed, but it can also have additional steps not included in the accompanying drawings. The process can correspond to a method, a function, a procedure, a subroutine, a subprogram, etc.
[0035] In structured light measurement technology, when acquiring point cloud data, due to the influence of equipment accuracy, operator experience and environmental factors, the diffraction of electromagnetic waves, changes in the surface properties of the measured object and the influence of the data stitching and registration operation process, some noise will inevitably appear in the point cloud data, and the sampling resolution is also different. Therefore, the original point cloud data obtained by the structured light system often contains a large number of hash points and isolated points, which requires point cloud noise reduction processing.
[0036] The point cloud denoising processing method of the present invention divides the neighborhood point cloud data of the point cloud to be processed into different sub-neighborhood point cloud data based on a predefined border cube, and performs bilateral filtering on these sub-neighborhood point cloud data to achieve outstanding edge preservation performance and high-precision point cloud denoising, and reduces the complexity of the method.
[0037] Reference Figure 1 The point cloud denoising method of the present invention is described in conjunction with a specific embodiment, and includes the following steps:
[0038] S1. Use a structured light system to obtain point cloud data of the object to be processed at different viewing angles, and use a point cloud stitching algorithm to stitch the point cloud data at different viewing angles to obtain original point cloud data;
[0039] S2, obtaining point cloud data, using the K nearest neighbor algorithm to obtain the neighborhood point cloud data of the point cloud to be filtered, and defining a neighborhood point cloud data set, expressed as N(p);
[0040] S3, define eight bounding cubes, refer to Figure 2, the index of each bounding box cube is defined as 1, 2, 3, 4, 5, 6, 7, 8 respectively;
[0041] Place eight bounding cubes together and set their common intersection as the point cloud to be filtered. Use the eight bounding cubes to divide the point cloud to be filtered into eight sub-neighborhood point clouds, represented as N i (p), that is, each bounding box contains part of the neighborhood point cloud, where the side length of the bounding box is greater than the radius of the neighborhood point cloud;
[0042] S4, performing bilateral filtering on the point cloud data contained in the eight bounding box cubes to obtain each new point cloud data;
[0043] set up is the unit normal vector of point cloud p, and its position is represented by l p , then the normal vector of the bilateral filter at the point cloud p to be processed can be expressed as:
[0044]
[0045] in, is the unit normal vector of the neighborhood point cloud of point cloud p; W C is a standard Gaussian filter, where W S is also a standard Gaussian filter, where
[0046] is the unit normal vector and The intensity difference between them is expressed as:
[0047] After bilateral filtering of each sub-neighborhood point cloud set, the new point cloud data obtained can be expressed as:
[0048]
[0049] in, Represents the sub-neighborhood point cloud location.
[0050] S5, using radius filtering to filter the point cloud to be filtered and eight new point cloud data;
[0051] Assume the radius filter center is the point cloud to be filtered, and the radius is R;
[0052] If the circle set by the radius filter contains only the point cloud to be filtered, and the number of point clouds in the circle is equal to 1, then the point cloud is removed, such as Figure 3a )
[0053] If the number of point clouds contained in the circle set by the radius filter is greater than 1, bilateral filtering is performed again on the point cloud contained in the circle to obtain the final filtering result. Figure 3b )
[0054] At this point, the denoising process of the point cloud is completed.
[0055] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any form. Any simple modification or equivalent change made to the above embodiment based on the technical essence of the present invention shall fall within the protection scope of the present invention.
Claims
1. A point cloud denoising method, characterized in that: The following steps are involved: A predefined border cube is used to divide a neighborhood point cloud into a plurality of sub-neighborhood point clouds with the point cloud to be filtered as the center; the method of dividing the neighborhood point cloud into a plurality of sub-neighborhood point clouds is as follows: eight border cubes are defined, and the eight border cubes are placed together, and their common intersection is the point cloud to be filtered; Perform bilateral filtering on each sub-neighborhood point cloud data to obtain a new point cloud; For the filtered point cloud and the new point cloud, radius filtering is performed with the filtered point cloud as the center to complete the denoising of the point cloud.
2. The point cloud denoising method according to claim 1, characterized in that: The method for obtaining the neighborhood point cloud data of the point cloud to be filtered is: obtaining the original point cloud data, and using the K nearest neighbor algorithm to obtain the neighborhood point cloud data of the point cloud to be filtered.
3. The point cloud denoising method according to claim 2, characterized in that: A structured light system is used to obtain point cloud data of the object to be processed at different viewing angles, and the point cloud data at different viewing angles are stitched together using a point cloud stitching algorithm to obtain the original point cloud data.
4. The point cloud denoising method according to claim 1, characterized in that: The side length of each bounding cube is larger than the radius of the neighborhood point cloud.
5. The point cloud denoising method according to claim 1, characterized in that: The method of using bilateral filtering to filter each sub-neighborhood point cloud data to obtain a new point cloud is: Assume that the point cloud to be filtered is p , p The neighborhood point cloud set is , each sub-neighborhood point cloud is ; Then bilateral filtering is used in point cloud p The normal vector at is: in, Point Cloud p The unit normal vector of , It is a point cloud p The unit normal vector of the neighborhood point cloud; is a standard Gaussian filter, , is a standard Gaussian filter, ; is the unit normal vector and The intensity difference between them is expressed as: ; After bilateral filtering of each sub-neighborhood point cloud set, the new point cloud data obtained can be expressed as: in, Represents the sub-neighborhood point cloud location.
6. The point cloud denoising method according to claim 1, characterized in that: When performing radius filtering on the point cloud to be filtered and the new point cloud, when the point cloud data in the circle is equal to 1, the point cloud is eliminated.
7. The point cloud denoising method according to claim 1, characterized in that: When radius filtering is performed on the to-be-filtered point cloud and the new point cloud, if the number of point clouds in the circle is greater than 1, bilateral filtering is performed again on the point cloud data in the circle.
Citation Information
Patent Citations
Filter method of nonuniform unorganized-point cloud data
CN103853840A
Rapid lossless filtering method for disordered 3D point cloud
CN113177897A
Improved Poisson surface reconstruction method based on nonlocal mean and bilateral filtering
CN108510591A
Improved guided filtering three-dimensional scattered point cloud rapid fairing method
CN111754421A
Statistical and bilateral filtering point cloud denoising method based on improved neighborhood search
CN114049267A
