Curvature entropy adaptive parameter-based point cloud filtering method

Abstract

The invention discloses a point cloud filtering method based on curvature entropy adaptive parameters, and belongs to the technical field of computer vision and three-dimensional data processing. The method comprises the following steps: firstly, for each target point in a point cloud, calculating a curvature representing a local geometrical shape of the target point; secondly, statistical binning is carried out on curvature values of the target point and all points in the neighborhood of the target point, and a discrete local curvature probability distribution histogram is constructed; then, the curvature entropy of the target point is calculated according to the probability distribution; thirdly, establishing a dynamic response model between the filtering parameters and the curvature entropy, and adaptively adjusting each parameter in a filtering weight function by using the calculated curvature entropy; and finally, performing weighted filtering on the target point by adopting the adaptively adjusted filtering parameters to complete filtering processing. According to the method, the curvature entropy based on statistical distribution is introduced to serve as a control variable of geometric features, point-by-point self-adaptive optimization of filtering parameters is achieved, and the protection capacity for fine features in a complex scene and the noise suppression effect are remarkably improved.

Description

Technical Field

[0001] This invention belongs to the field of computer vision and 3D data processing technology, specifically relating to a point cloud filtering method based on curvature entropy adaptive parameters. Background Technology

[0002] Point clouds, as an important digital representation of the three-dimensional world, are widely used in reverse engineering, autonomous driving, cultural relic preservation, and robot navigation. However, during the acquisition process, raw point cloud data often contains a large amount of noise due to factors such as equipment precision and environmental interference, which seriously affects the accuracy of subsequent advanced processing such as 3D reconstruction, feature extraction, and target recognition. Therefore, point cloud filtering is a crucial step in the point cloud data preprocessing workflow.

[0003] Existing point cloud filtering algorithms, such as bilateral filtering and moving least squares, typically employ a weighted average framework to update point positions. The performance of these algorithms largely depends on the selection of their filtering parameters, such as neighborhood size, spatial distance weights, and normal vector deviation weights. In traditional methods, these parameters are usually set as globally fixed constants, requiring users to manually adjust them based on experience.

[0004] However, this fixed-parameter strategy has inherent flaws: poor adaptability. Different regions of a point cloud have vastly different geometric features. For example, a larger filtering scale is needed in flat regions to effectively remove noise, while a smaller filtering scale is needed in high-frequency feature regions such as edges and corners to avoid feature blurring. Fixed global parameters cannot simultaneously meet these two contradictory needs, often resulting in incomplete noise removal in flat regions or excessive smoothing in feature regions, destroying key geometric details. It is also inefficient and highly subjective. The manual parameter tuning process is time-consuming and labor-intensive, and the tuning results heavily depend on the operator's professional knowledge and experience, lacking objectivity and reproducibility. For large-scale or geometrically complex point clouds, finding a globally optimal set of parameters is almost impossible.

[0005] Therefore, designing a point cloud filtering algorithm that can accurately perceive the complexity of the local geometric environment and automatically and finely adjust the filtering parameters accordingly to achieve the best balance between noise suppression and feature preservation is a pressing technical challenge. Based on this, a point cloud filtering method based on curvature entropy adaptive parameters is proposed. Summary of the Invention

[0006] The purpose of this invention is to provide a point cloud filtering method that can accurately perceive the local geometric structure features of point clouds and adaptively adjust filtering parameters point-by-point and in multiple dimensions, while effectively suppressing the influence of outliers. This addresses the technical problems existing in current point cloud filtering algorithms, such as parameter fixation, insufficient feature recognition accuracy, and outlier interference. It is suitable for denoising, feature enhancement, and geometric structure optimization of 3D scanned point cloud data, and is widely used in fields such as autonomous driving, reverse engineering, and human-computer interaction.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: A point cloud filtering method based on curvature entropy adaptive parameters, the method comprising the following steps: S1 calculates the curvature value for each point in the input point cloud; where for the target point... A local covariance matrix is ​​constructed by determining its neighborhood point set, and the covariance matrix is ​​then decomposed into three eigenvalues. And then according to the formula Its curvature value was calculated; S2, for the target point and its neighboring point set, the curvature value range is divided into multiple non-overlapping intervals. The number of points in each interval is counted to construct a local curvature probability distribution, and then the information entropy formula is applied. Calculate the curvature entropy of the target point ,in Let be the probability value of the k-th interval; S3, Establish curvature entropy Based on the functional relationship between multiple filter parameters in the filter weight function, and according to the curvature entropy, each filter parameter is adaptively adjusted to obtain a personalized filter parameter set for the target point. S4. Construct a filter weight function based on the personalized filter parameter set, and use this function to perform a weighted average filter on the neighborhood points of the target point to calculate the coordinates of the filtered point. This enables point-by-point adaptive filtering.

[0008] Furthermore, the process of constructing the local curvature probability distribution specifically includes: S2.1, Pre-define the global range of curvature values, and divide this range into... A number of non-overlapping curvature intervals; S2.2, for the target point And its neighborhood set, count the number of points falling into each curvature interval; S2.3, divide the number of points counted within each curvature interval by the total number of points in the neighborhood point set to obtain the probability value corresponding to that interval. , The probability values ​​of all intervals together constitute the local curvature probability distribution of the target point.

[0009] Furthermore, the filtering parameters include spatial distance weighting parameters. Normal vector deviation weight parameter and entropy enhancement factor .

[0010] Furthermore, the specific form of the filter weight function is as follows: ; in, These are the coordinates of the target point and its neighboring points, respectively. These are the normal vectors of the target point and its neighboring points, respectively. For target points respectively Spatial distance weight parameter, normal vector deviation weight parameter, and entropy enhancement factor, It is the outlier constraint factor.

[0011] Furthermore, the spatial distance weighting parameter The adaptive adjustment method is as follows: curvature entropy The value is compared with a preset entropy threshold, and adjustments are made in segments based on the comparison results: When curvature entropy When the entropy exceeds a preset threshold, reduce When curvature entropy When the entropy threshold is lower than or equal to the preset threshold, increase ,and The value of curvature They are inversely proportional.

[0012] Furthermore, the normal vector deviation weight parameter The adaptive adjustment method is as follows: Calculate the local curvature entropy gradient of the target point. Its value is the curvature entropy of the target point. Its curvature entropy at points in its neighborhood The maximum value of the absolute value of the difference; The local curvature entropy gradient is input into a monotonically increasing function model and mapped to obtain... The value of .

[0013] Furthermore, the monotonically increasing function model is the Sigmoid function.

[0014] Furthermore, the entropy enhancement factor The adaptive adjustment method is as follows: Entropy enhancement factor Set as relative to the curvature entropy Variables with negative correlation; Based on the aforementioned negative correlation, it increases in the low curvature entropy region. The value decreases in the region of high curvature entropy. value.

[0015] Furthermore, the outlier constraint factor Here is a distance-related indicator function: ; in, The distance threshold is set based on the overall average distance of the input point cloud, where d represents the average distance from the target point to its neighboring points.

[0016] Furthermore, in S4, the filtered point coordinates The calculation formula is as follows: ; in, The coordinates of the neighborhood points of the target point. For target point With neighboring points The filter weights between them For target point The neighborhood point set.

[0017] The beneficial effects of this invention are: This invention addresses the technical problems of existing point cloud filtering algorithms, such as parameter fixation, insufficient feature recognition accuracy, and outlier interference. By introducing curvature entropy and its gradient as control variables, this method can accurately identify flat regions, feature regions, and transition boundaries between them. It also performs refined and multi-dimensional adaptive adjustment of filtering parameters, thereby effectively filtering out noise while preserving high-frequency geometric details such as edges and corners of the point cloud to the maximum extent. At the same time, the introduction of outlier constraint factors can effectively resist the interference of strong noise points, and stable and reliable filtering results can be obtained even under conditions of high noise levels or missing data. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating a point cloud filtering method based on curvature entropy adaptive parameters. Figure 2 This is the filtering result of the algorithm shown in this embodiment. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present invention and not intended to limit the scope of the invention.

[0020] like Figure 1 As shown in the figure, this embodiment of the invention provides a point cloud filtering method based on curvature entropy adaptive parameters, and its various steps will be described in detail below.

[0021] Input: Origin Point Cloud Collection ,in Let i be the three-dimensional coordinates of point i, and N be the total number of points in the cloud.

[0022] Output: Filtered point cloud .

[0023] Step S1, Local curvature calculation For Dianyunji Each target point in Perform the following operations to calculate its local curvature value: (1.1) Neighborhood search The target point is determined using the K-nearest neighbor algorithm or the radius search method. neighborhood point set neighborhood size These are preset parameters used to control the calculation range of local geometric features.

[0024] (1.2) Construction of covariance matrix Calculate the centroid of the neighborhood point set: Then construct based on the centroid Local covariance matrix : in, For target point neighborhood points, For target point The centroid of the neighborhood set of points, with superscript T representing the transpose.

[0025] (1.3) Eigenvalue decomposition For covariance matrix Eigenvalue decomposition yields three non-negative eigenvalues. and its corresponding eigenvectors, these three eigenvalues The variances of the neighborhood points in the three principal directions are described respectively.

[0026] (1.4) Calculation of point curvature Calculate the target point using the following formula. curvature : This curvature value A larger value indicates a higher degree of curvature or irregularity in the local area where that point is located. Furthermore, for... All points in the neighborhood Repeat this process or use existing calculation results to obtain their curvature. This is used for subsequent curvature entropy calculations.

[0027] Step S2, Local curvature entropy calculation For the target point and its neighborhood Perform the following operations to calculate its curvature entropy. , used to quantify local geometric complexity: (2.1) Division of curvature intervals A global range of curvature values ​​is predefined; in this embodiment, it is taken as... Divide this range into equal intervals. Each non-overlapping bin forms a discrete statistical interval for curvature values. In this embodiment, we take... The curvature range is evenly divided into 32 equal-width intervals.

[0028] (2.2) Curvature distribution statistics Traversing the neighborhood point set Calculate the curvature value at each point. Falling into the above Which of the intervals, record the number of points in each interval. ( ).

[0029] (2.3) Calculation of probability distribution Based on the statistical results, calculate the probability value for each interval. : in, This represents the total number of points in the neighborhood point set. The probability values ​​for all intervals. The local curvature probability distribution that together constitutes the target point.

[0030] (2.4) Calculation of curvature entropy Based on the information entropy theory, calculate the target point. curvature entropy : To avoid hour Meaningless, the agreement should hour, Curvature entropy The range of values ​​is The larger the value, the more complex and inconsistent the geometry within the neighborhood.

[0031] Step S3, adaptive adjustment of filter parameters Based on the calculated curvature entropy and point curvature , for target point Generate a set of personalized filter parameters This step uses a dynamic response model to adaptively adjust parameters point by point, balancing noise suppression and feature preservation. The specific adjustment process is as follows: (3.1) Spatial distance weighting parameters Adjustment A piecewise function is used for adjustment, and an entropy threshold is set. In this embodiment, Used to distinguish between high-entropy and low-entropy regions, when curvature entropy When the entropy exceeds a preset threshold, reduce By shrinking the filtering neighborhood, potential complex features are protected; when curvature entropy When the entropy is less than or equal to the stated entropy threshold, based on the point curvature The reciprocal relationship increases To expand the filtering range in flat or structurally simple regions.

[0032] Specifically, if This is a high-entropy region, which may correspond to feature edges or noise points. Therefore, it's necessary to shrink the filtering neighborhood to preserve details. The adjustment formula is: in, Based on the basic space parameters, The adjustment coefficient controls the rate at which the entropy value decays with respect to the spatial weights.

[0033] like This is a low-entropy region, which corresponds to a flat or simple structure. The filtering range can be expanded for better noise reduction. The adjustment formula is: in, As another basic parameter, It is a very small normal number to prevent curvature The denominator is zero.

[0034] By segmenting the adjustment, it is possible to achieve an adaptive capability that protects features in complex areas and enhances noise reduction in flat areas.

[0035] (3.2) Normal vector deviation weight parameter Adjustment This parameter controls the impact of normal vector consistency on the filter weights, primarily responding to feature boundaries, and its adjustment focuses on feature boundary identification.

[0036] First, calculate the local curvature entropy gradient of the target point. To quantify the degree of change in geometric complexity within the neighborhood: The larger the gradient value, the more likely the target point is located in the feature boundary region.

[0037] Then, taking the local curvature entropy gradient as input, the entropy gradient is mapped to the parameter values ​​using a monotonically increasing sigmoid function. : in, These are the minimum and maximum ranges of the normal parameter, respectively. The adjustment coefficient controls the steepness of the function. It is the center threshold of the gradient response. In feature boundary regions with large curvature entropy gradients, The value of is also increased accordingly to improve tolerance to changes in the normal direction.

[0038] (3.3) Entropy enhancement factor Adjustment This factor is used to modulate the overall strength of the filter weights, in relation to curvature entropy. A negative correlation indicates the confidence level in the filtering result. The adjustment formula is: in, This is the maximum enhancement factor, which is set to 1.0 in this embodiment. It is an adjustment index used to control the decay rate. The lower, The larger the entropy enhancement factor, the better. In regions with low curvature entropy, representing highly consistent local geometric features, the entropy enhancement factor is increased. The value of is adjusted to enhance the determinism of filtering in that region; while in regions with high curvature entropy, representing complex or chaotic local geometric features, the value is reduced to mitigate the impact of uncertainty.

[0039] Step S4, weighted filtering Based on the personalized filter parameter set obtained in step S3, the target point The weighted filtering process is as follows: (4.1) Calculation of filter weights For the target point and each point in its neighborhood Calculate its filter weights : in, These are the coordinates of the target point and its neighboring points, respectively. This represents the Euclidean distance between two points. Let be the normal vectors of the target point and its neighboring points, respectively. Here, the normal vector... It is obtained by finding the minimum eigenvalue of the covariance matrix in step S1. The corresponding feature vectors are obtained. This indicates the degree of difference between the normal vectors of two points. For target points respectively Spatial distance weight parameter, normal vector deviation weight parameter, and entropy enhancement factor, It is the outlier constraint factor.

[0040] In this embodiment, the outlier constraint factor Here is a distance-related indicator function, calculated as follows: in, for Average distance to neighboring points In this embodiment, a distance threshold related to the average distance of the input point cloud is set to 1.5 times the average distance between points in the entire point cloud. When the average distance from a point to its neighbors is greater than the threshold distance, the point is considered an outlier and is removed. Conversely, when the average distance from a point to its neighbors is less than the threshold distance, the point is retained. This factor effectively resets the weight of outliers to 0.

[0041] (4.2) Point coordinate update Calculate the coordinates of the new point after filtering. : If the denominator is zero (i.e., all weights are zero), then the original position remains unchanged. .

[0042] Point-to-point gathering Repeat steps S1-S4 above for all points to obtain the final filtered point cloud. .

[0043] like Figure 2 The image shows the actual point cloud filtering results in this embodiment. Figure 2 In this context, 'a' represents the input noise point cloud. Figure 2 In this context, 'b' represents the point cloud after filtering by the algorithm. Figure 2 The images above and below 'c' are magnified views of the point cloud before and after filtering, respectively. It can be seen that the noise in the point cloud is significantly removed by the algorithm filtering, while the edge details are well adjusted and preserved.

[0044] The specific embodiments of the present invention have been described above with reference to the accompanying drawings. However, these descriptions should not be construed as limiting the scope of the present invention. The scope of protection of the present invention is defined by the appended claims. Any modifications based on the claims of the present invention are within the scope of protection of the present invention.

Claims

1. A point cloud filtering method based on curvature entropy adaptive parameters, characterized in that, The method includes the following steps: S1 calculates the curvature value for each point in the input point cloud; where for the target point... A local covariance matrix is ​​constructed by determining its neighborhood point set, and the covariance matrix is ​​then decomposed into three eigenvalues. And then according to the formula Its curvature value was calculated; S2, for the target point and its neighboring point set, the curvature value range is divided into multiple non-overlapping intervals. The number of points in each interval is counted to construct a local curvature probability distribution, and then the information entropy formula is applied. Calculate the curvature entropy of the target point ,in Let be the probability value of the k-th interval; S3, Establish curvature entropy Based on the functional relationship between multiple filter parameters in the filter weight function, and according to the curvature entropy, each filter parameter is adaptively adjusted to obtain a personalized filter parameter set for the target point. S4. Construct a filter weight function based on the personalized filter parameter set, and use this function to perform a weighted average filter on the neighborhood points of the target point to calculate the coordinates of the filtered point. This enables point-by-point adaptive filtering.

2. The point cloud filtering method based on curvature entropy adaptive parameters according to claim 1, characterized in that, In S2, the process of constructing the local curvature probability distribution specifically includes: S2.1, Pre-define the global range of curvature values, and divide this range into... A number of non-overlapping curvature intervals; S2.2, for the target point And its neighborhood set, count the number of points falling into each curvature interval; S2.3, divide the number of points counted within each curvature interval by the total number of points in the neighborhood point set to obtain the probability value corresponding to that interval. , The probability values ​​of all intervals together constitute the local curvature probability distribution of the target point.

3. The point cloud filtering method based on curvature entropy adaptive parameters according to claim 1, characterized in that, In S3, the filtering parameters include spatial distance weighting parameters. Normal vector deviation weight parameter and entropy enhancement factor .

4. The point cloud filtering method based on curvature entropy adaptive parameters according to claim 3, characterized in that, In S4, the specific form of the filter weight function is as follows: ； in, These are the coordinates of the target point and its neighboring points, respectively. These are the normal vectors of the target point and its neighboring points, respectively. For target points respectively Spatial distance weight parameter, normal vector deviation weight parameter, and entropy enhancement factor, It is the outlier constraint factor.

5. The point cloud filtering method based on curvature entropy adaptive parameters according to claim 3, characterized in that, The spatial distance weight parameter The adaptive adjustment method is as follows: curvature entropy The value is compared with a preset entropy threshold, and adjustments are made in segments based on the comparison results: When curvature entropy When the entropy exceeds a preset threshold, reduce When curvature entropy When the entropy threshold is lower than or equal to the preset threshold, increase ,and The value of curvature They are inversely proportional.

6. The point cloud filtering method based on curvature entropy adaptive parameters according to claim 3, characterized in that, The normal vector deviation weight parameter The adaptive adjustment method is as follows: Calculate the local curvature entropy gradient of the target point. Its value is the curvature entropy of the target point. Its curvature entropy at points in its neighborhood The maximum value of the absolute value of the difference; The local curvature entropy gradient is input into a monotonically increasing function model and mapped to obtain... The value of .

7. The point cloud filtering method based on curvature entropy adaptive parameters according to claim 6, characterized in that, The monotonically increasing function model is the Sigmoid function.

8. The point cloud filtering method based on curvature entropy adaptive parameters according to claim 3, characterized in that, The entropy enhancement factor The adaptive adjustment method is as follows: Entropy enhancement factor Set as relative to the curvature entropy Variables with negative correlation; Based on the aforementioned negative correlation, it increases in the low curvature entropy region. The value decreases in the region of high curvature entropy. value.

9. The point cloud filtering method based on curvature entropy adaptive parameters according to claim 4, characterized in that, The outlier constraint factor Here is a distance-related indicator function: ； in, The distance threshold is set based on the overall average distance of the input point cloud, where d represents the average distance from the target point to its neighboring points.

10. The point cloud filtering method based on curvature entropy adaptive parameters according to claim 1, characterized in that, In S4, the filtered point coordinates The calculation formula is as follows: ； in, The coordinates of the neighborhood points of the target point. For target point With neighboring points The filter weights between them For target point The neighborhood point set.

Images