An unmanned aerial vehicle-borne laser radar point cloud data denoising and outlier detection method

Abstract

The application discloses a kind of unmanned aerial vehicle laser radar point cloud data denoising and outlier detection method, it is related to point cloud data denoising technical field, comprising the following steps: based on unmanned aerial vehicle laser radar acquisition target's point cloud data, and pre-division processing is carried out to point cloud data, obtain first point cloud data;According to first point cloud data, radius acquisition processing is carried out, and obtains denoising radius data;Based on first point cloud data, corresponding final point number threshold value is obtained, and according to denoising radius data, first point cloud denoising processing is carried out, and second point cloud data is obtained;Based on second point cloud data, point cloud outlier detection processing is carried out, and third point cloud data is obtained;The application is used to solve the problem that when current point cloud data denoising technology is based on radius filtering method to laser radar point cloud data denoising, cannot automatically set the best radius range and quantity threshold according to point cloud density, accurately denoising the problem of point cloud data of uneven density.

Description

Technical Field

[0001] This invention relates to the field of point cloud data denoising technology, specifically a method for denoising and outlier detection of point cloud data from UAV-borne lidar. Background Technology

[0002] Point cloud data denoising technology refers to the process of identifying and removing noise points from point cloud data and retaining real and valid points through algorithms or processing methods. Its core goal is to improve the quality of point cloud data so that it can more accurately reflect the real geometric structure of objects or scenes, laying the foundation for subsequent tasks such as 3D modeling, surface reconstruction, and target recognition.

[0003] Existing point cloud data denoising techniques based on radius filtering for LiDAR point cloud data often rely on manually setting a fixed radius range and a threshold for the number of neighboring points to determine whether a point is noise. However, LiDAR point cloud data exhibits significant differences in point cloud density across different scenarios. A fixed radius range can lead to normal points in low-density areas being easily misidentified as noise, while noise points in high-density areas may be retained. Furthermore, these techniques are unsuitable for point cloud data with uneven density, and the point cloud density may differ across different parts of the same object. Fixed parameters can also result in inconsistent denoising effects across different areas of the same object. Additionally, they cannot adapt to changes in target scale. For example, patent application CN116843563A discloses a point cloud denoising method that uses a manually set radius range and a threshold for the number of neighboring points for radius filtering, resulting in low accuracy. Therefore, existing point cloud data denoising techniques based on radius filtering for LiDAR point cloud data cannot automatically set the optimal radius range and threshold based on point cloud density to accurately denoise point cloud data with uneven density. Summary of the Invention

[0004] This invention aims to at least partially solve one of the technical problems in the prior art. It obtains first point cloud data by acquiring point cloud data of a target based on UAV-borne LiDAR and pre-dividing the point cloud data; performs radius acquisition processing based on the first point cloud data; obtains the corresponding final point count threshold based on the first point cloud data and performs first point cloud denoising processing to obtain second point cloud data; and performs point cloud outlier detection processing based on the second point cloud data to obtain third point cloud data. This solves the problem that existing point cloud data denoising technologies based on radius filtering cannot automatically set the optimal radius range and number threshold according to the point cloud density when denoising LiDAR point cloud data, thus failing to accurately denoise point cloud data with uneven density.

[0005] To achieve the above objectives, this application provides a method for denoising and outlier detection of UAV-borne lidar point cloud data, comprising the following steps:

[0006] The first point cloud data is obtained by collecting point cloud data of the target using UAV-borne lidar and pre-dividing the point cloud data.

[0007] Based on the first point cloud data, the radius is processed to obtain the corresponding optimal denoising radius, which is recorded as the denoising radius data.

[0008] Based on the first point cloud data, the corresponding final point number threshold is obtained, and the first point cloud is denoised according to the denoising radius data to obtain the second point cloud data;

[0009] Point cloud outlier detection processing is performed based on the second point cloud data to obtain the third point cloud data.

[0010] Furthermore, the process of acquiring point cloud data of the target object based on UAV-borne lidar and pre-segmenting the point cloud data to obtain the first point cloud data includes the following sub-steps:

[0011] For the target whose point cloud data is to be collected, it is denoted as the first target. The point cloud data of the first target is collected using the UAV's airborne lidar and is denoted as the raw point cloud data.

[0012] Establish a three-dimensional spatial coordinate system and obtain the coordinates of all points in the original point cloud data in the three-dimensional spatial coordinate system, which are denoted as point cloud coordinate data.

[0013] Furthermore, the process of acquiring point cloud data of the target object based on UAV-borne lidar and pre-segmenting the point cloud data to obtain the first point cloud data also includes the following sub-steps:

[0014] Obtain the extreme values ​​of the horizontal axis, vertical axis, and vertical axis coordinates in the point cloud coordinate data respectively, and construct the minimum bounding cube of the original point cloud data based on the extreme values ​​of the horizontal axis, vertical axis, and vertical axis coordinates, denoted as the first cube;

[0015] The first cube is evenly divided into k1 sub-cubes, denoted as the second cube. After this is completed, the first point cloud data is obtained, where k1 is the set number of divisions.

[0016] Further, the radius acquisition process is performed based on the first point cloud data to obtain the corresponding optimal denoising radius, denoised as the denoising radius data. This includes the following sub-steps:

[0017] Set the initial neighborhood radius to RE and the maximum neighborhood radius to RM; for any second cube, denote it as the second cubic region; for any point in the second cubic region, denote it as the first center point;

[0018] The spherical region with the first center point as the center and RE as the radius is denoted as the initial neighborhood of the first center point;

[0019] Obtain the distances from all points in the initial neighborhood of the first center point to the first center point, denoted as the neighborhood distance sequence; and calculate the average and variance of all distances in the neighborhood distance sequence, denoted as the corresponding mean PE and variance PB of the first center point in order.

[0020] Further, the radius acquisition process based on the first point cloud data to obtain the corresponding optimal denoising radius, denoted as the denoising radius data, also includes the following sub-steps:

[0021] Repeatedly obtain the variance of all points within the second cubic region, and calculate the maximum value of the corresponding variance, denoted as PM; calculate the optimal denoising radius corresponding to the first center point according to the first formula; the first formula is as follows: , where RL represents the excellent denoising radius;

[0022] Repeatedly obtain the excellent denoised radii corresponding to all points in the second cubic region, and arrange them in ascending order, denoted as the excellent radius sequence; calculate the average of the k2th percentile to the k3th percentile in the excellent radius sequence, and denoted as the suspected optimal radius corresponding to the second cubic region; where k2 and k3 are the set percentiles;

[0023] If the suspected optimal radius corresponding to the second cubic region is less than the maximum neighborhood radius, then mark the suspected optimal radius corresponding to the second cubic region as the optimal denoising radius of the second cubic region; otherwise, mark the maximum neighborhood radius as the optimal denoising radius of the second cubic region.

[0024] Repeatedly obtain the optimal denoising radius for all second cubes to get the denoising radius data.

[0025] Further, based on the first point cloud data, the corresponding final point count threshold is obtained, and the first point cloud is denoised according to the denoising radius data to obtain the second point cloud data, including the following sub-steps:

[0026] The second cubic region is evenly divided into k4 sub-cubes, denoted as the third cube. Any third cube is denoted as the third cubic region, where k4 is the number of divisions set.

[0027] For any point in the third cubic region, it is denoted as the second center point; the spherical region with the second center point as the center and the optimal noise reduction radius corresponding to the second center point as the radius is denoted as the initial neighborhood of the second center point.

[0028] Count the number of points in the initial neighborhood of the second center point, denoted as the neighborhood number NA; and calculate the neighborhood density corresponding to the second center point according to the second formula, which is as follows: Where DM represents neighborhood density and VR represents the optimal denoising radius;

[0029] Repeatedly obtain the number of neighborhood points and the neighborhood density corresponding to all points in the third cubic region, and arrange them in ascending order, denoted as the neighborhood point number sequence and the neighborhood density sequence, respectively. Obtain the maximum value and median in the neighborhood point number sequence, and denot them as NM and NZ in order; obtain the median of the neighborhood density sequence, denoted as DZ.

[0030] Furthermore, obtaining the corresponding final point count threshold based on the first point cloud data, and performing denoising processing on the first point cloud according to the denoising radius data to obtain the second point cloud data also includes the following sub-steps:

[0031] Set the maximum number of points threshold to DS0 and the minimum number of points threshold to DS1; calculate the first number of points coefficient XE1, the second number of points coefficient XE2 and the third density coefficient XE3 respectively, where XE1=min(DS0, k5*NM), XE2=max(DS1, k6*NZ), XE3=1 / (k7*DZ); where k5, k6 and k7 are proportional coefficients;

[0032] The initial point threshold corresponding to the second center point is calculated using the third formula, which is as follows: , where CY represents the initial point threshold; the initial point threshold for repeatedly obtaining all points in the third cubic region;

[0033] Calculate the average of the initial point count thresholds for all points in the initial neighborhood of the second center point, excluding the second center point itself, and denote this as the average of other thresholds CP. Then, calculate the final point count threshold for the second center point using the fourth formula, which is as follows: Where CZ represents the final point threshold, and q1 and q2 are weight coefficients; the final point threshold corresponding to all points in the first cube is obtained repeatedly.

[0034] Furthermore, obtaining the corresponding final point count threshold based on the first point cloud data, and performing denoising processing on the first point cloud according to the denoising radius data to obtain the second point cloud data also includes the following sub-steps:

[0035] For any point in the first cube, denote it as an arbitrary target point; denote the spherical region with the arbitrary target point as the center and the optimal denoising radius corresponding to the arbitrary target point as the radius as the arbitrary target neighborhood. If the number of points in the arbitrary target neighborhood is less than the final number threshold corresponding to the arbitrary target point, then mark the arbitrary target point as a noise point; repeat the marking of all points in the first cube and remove the noise points to obtain the second point cloud data.

[0036] Furthermore, the process of detecting outliers in the second point cloud data to obtain the third point cloud data includes the following sub-steps:

[0037] The first cube corresponding to the second point cloud data is evenly divided into k8 sub-cubes, and any one of the cubes is denoted as the fourth cube region; any point in the fourth cube region is denoted as the first target point; where k8 is the set number of divisions.

[0038] Obtain the g1 points within the fourth cubic region that are closest to the first target point, and denote them as the first neighboring points. Calculate the square of the Euclidean distance between each of the first neighboring points and the first target point, and sum them to obtain the sum of squared distances to the first target point, denoted as FS.

[0039] Repeatedly obtain the sum of squared distances of all points within the fourth cubic region, and calculate the average of the sums of squared distances. This average is denoted as FP.

[0040] The ratio of the squared distance to the first target point is calculated using the fifth formula, which is as follows: , where GS represents the distance squared ratio; the distance squared ratio of all points in the fourth cubic region is repeatedly obtained.

[0041] Furthermore, the process of detecting outliers in the second point cloud data to obtain the third point cloud data also includes the following sub-steps:

[0042] Obtain the squared ratio of the distance between the first target point and its corresponding first neighboring point, denoted as the squared ratio sequence. Calculate the mean and standard deviation of the squared ratio sequence, and denot them as GP and GB respectively in order.

[0043] The dynamic threshold GY of the first target point is calculated based on GP and GB, where GY = GP + g2 * GB, and g2 is the set proportional coefficient. If the squared distance ratio of the first target point is greater than the dynamic threshold GY of the first target point, the first target point is marked as an abnormal point; otherwise, the first target point is marked as a normal point.

[0044] Repeat the detection and labeling process for all points in the second point cloud data to obtain the third point cloud data.

[0045] The beneficial effects of this invention are as follows: This invention is based on point cloud data of a target collected by an UAV-borne lidar, and pre-divides the point cloud data to obtain first point cloud data; radius acquisition processing is performed on the first point cloud data to obtain the corresponding optimal denoising radius, denoised as denoising radius data; the corresponding final point number threshold is obtained based on the first point cloud data, and the first point cloud is denoised based on the denoising radius data to obtain second point cloud data; point cloud outlier detection processing is performed on the second point cloud data to obtain third point cloud data; when using the radius filtering method to denoise lidar point cloud data, the optimal radius range and number threshold can be automatically set according to the point cloud density to accurately denoise point cloud data with uneven density;

[0046] This invention independently calculates the optimal denoising radius for each second cube's point cloud, ensuring that the radius parameter within the same region is only affected by local point cloud features, thus improving parameter specificity; it makes the radius positively correlated with the local dispersion, enhancing robustness to sparse regions and outliers, and improving denoising accuracy in complex scenes; it independently calculates the point count threshold for each third cube, which can adapt to the problem of density differences in different parts of the same region, achieving micro-scale adaptation; and it calculates the neighborhood density based on the corresponding optimal denoising radius, which, compared to the traditional threshold that only relies on the number of points, can more accurately distinguish between "true sparsity" and "false sparsity," thus improving denoising accuracy. Attached Figure Description

[0047] Figure 1 This is a flowchart illustrating the steps of the method of the present invention;

[0048] Figure 2 This is a flowchart of the radius acquisition process of the present invention;

[0049] Figure 3 This is a schematic diagram of the initial neighborhood of the first center point in this invention;

[0050] Figure 4 This is a schematic diagram of the electronic device of the present invention. Detailed Implementation

[0051] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0052] Example 1, please refer to Figure 1 As shown, this application provides a method for denoising and outlier detection of UAV-borne lidar point cloud data, including the following steps:

[0053] Step S1 involves collecting point cloud data of the target based on the UAV-borne lidar and pre-segmenting the point cloud data to obtain the first point cloud data. Step S1 includes the following sub-steps:

[0054] Step S101: For the target whose point cloud data is to be collected, denoted as the first target, the point cloud data of the first target is collected using the UAV's airborne lidar and denoted as the raw point cloud data.

[0055] Step S102: Establish a three-dimensional spatial coordinate system and obtain the coordinates of all points in the original point cloud data in the three-dimensional spatial coordinate system, which are recorded as point cloud coordinate data;

[0056] Step S103: Obtain the extreme values ​​of the horizontal axis coordinate, the vertical axis coordinate, and the vertical axis coordinate in the point cloud coordinate data respectively, and construct the minimum bounding cube of the original point cloud data based on the extreme values ​​of the horizontal axis coordinate, the vertical axis coordinate, and the vertical axis coordinate, denoted as the first cube; that is, establish the smallest volume cube that can enclose all the original point cloud data, denoted as the first cube.

[0057] Step S104: Divide the first cube evenly into k1 sub-cubes, denoted as the second cube. After completion, the first point cloud data is obtained, where k1 is the set number of divisions. k1 can be set according to the actual application scenario. LiDAR point clouds often exhibit uneven density in space. A globally uniform parameter cannot take into account the needs of different areas. Therefore, by dividing into sub-cubes, the point cloud is divided into multiple regions. Parameters are set according to the features of the point cloud in each region to improve the accuracy of noise reduction.

[0058] In the specific implementation process, for the establishment of a three-dimensional spatial coordinate system, the original point cloud data is generally placed in an area where the horizontal axis, vertical axis and vertical axis are all positive, which facilitates subsequent processing. Alternatively, the sensor coordinate system of the UAV-borne lidar itself can be used for subsequent processing.

[0059] Step S2 involves performing radius acquisition processing based on the first point cloud data to obtain the corresponding optimal denoising radius, which is denoised as the denoising radius data. Step S2 includes the following sub-steps:

[0060] For step S201, please refer to... Figure 2 As shown, the initial neighborhood radius is set to RE, and the maximum neighborhood radius is set to RM. For any second cube, it is denoted as the second cubic region. For any point in the second cubic region, it is denoted as the first center point. RE and RM can be set according to the actual application scenario. Generally, RE should be slightly larger or smaller than the average spacing of the point cloud to ensure that the initial neighborhood of each point contains enough neighboring points and avoid statistical failure due to the neighborhood being too small. RM should be smaller than the feature size of the smallest target in the actual scene to avoid the neighborhood being too large and containing points of irrelevant targets, which may cause noise to be misjudged as normal points.

[0061] For step S202, please refer to... Figure 3 As shown, the spherical region with the first center point as the center and RE as the radius is denoted as the initial neighborhood of the first center point;

[0062] Step S203: Obtain the distances from all points in the initial neighborhood of the first center point to the first center point, denoted as the neighborhood distance sequence; and calculate the average and variance of all distances in the neighborhood distance sequence, denoted as the corresponding mean PE and variance PB of the first center point in order; the variance PB represents the dispersion of points in the initial neighborhood. The larger PB is, the more dispersed the distribution of points in the neighborhood, and the higher the dispersion; the smaller PB is, the more concentrated the point cloud distribution, and the lower the dispersion.

[0063] Step S204: Repeatedly obtain the corresponding variances of all points within the second cubic region, and calculate the maximum value of the corresponding variances, denoted as PM; calculate the optimal denoising radius corresponding to the first center point according to the first formula; the first formula is as follows: , where RL represents the excellent denoising radius; when the corresponding mean of a point is 0, the radius takes the minimum value, and when the corresponding mean is PM, the radius takes the maximum value. In this way, the radius is adaptively changed with the dispersion of the neighborhood.

[0064] Step S205: Repeatedly obtain the excellent denoised radii corresponding to all points in the second cubic region, and arrange them in ascending order, denoted as the excellent radius sequence; calculate the average of the k2th percentile to the k3th percentile in the excellent radius sequence, denoted as the suspected optimal radius corresponding to the second cubic region; where k2 and k3 are set percentiles; in this embodiment, k2 is 15 and k3 is 85, that is, the average of the middle 70% of the radii in the excellent radius sequence is taken as the suspected optimal radius; through percentile screening, eliminate extreme radii that are too small or too large, and retain reasonable radius ranges that reflect the characteristics of most points; take the average of the interval as the suspected optimal radius, so that the radius parameters within the same sub-cube tend to be consistent, avoid drastic radius fluctuations caused by single-point noise, and ensure uniform denoising intensity;

[0065] Step S206: If the suspected optimal radius corresponding to the second cubic region is less than the maximum neighborhood radius, then mark the suspected optimal radius corresponding to the second cubic region as the optimal denoising radius of the second cubic region; otherwise, mark the maximum neighborhood radius as the optimal denoising radius of the second cubic region. RM serves as a scene-related maximum radius limit to prevent the algorithm from generating an excessively magnified radius and to avoid normal points being mistakenly deleted because their neighborhood contains irrelevant points.

[0066] Step S207: Repeatedly obtain the optimal denoising radius corresponding to all second cubes to obtain denoising radius data;

[0067] In the specific implementation process, the first cube is divided into multiple second cubes, and the optimal denoising radius is calculated independently based on the point cloud of each second cube. This achieves the combination of denoising radius with local point cloud density and dispersion. Compared with the traditional fixed radius method, it improves the adaptability, noise resistance robustness and parameter specificity, laying a precise parameter foundation for subsequent denoising.

[0068] Step S3: Obtain the corresponding final point count threshold based on the first point cloud data, and perform denoising processing on the first point cloud according to the denoising radius data to obtain the second point cloud data; Step S3 includes the following sub-steps:

[0069] Step S301: Divide the second cubic region evenly into k4 sub-cubes, denoted as the third cube. Any third cube is denoted as the third cubic region, where k4 is the set number of divisions. k4 can be set according to the actual application scenario. Smaller third cubes enable fine recognition of local point cloud and avoid density calculation distortion caused by excessively large factor regions.

[0070] Step S302: For any point in the third cubic region, it is denoted as the second center point; the spherical region with the second center point as the center and the optimal denoising radius corresponding to the second center point as the radius is denoted as the initial neighborhood of the second center point.

[0071] Step S303: Count the number of points in the initial neighborhood of the second center point, denoted as the neighborhood point count NA; and calculate the neighborhood density corresponding to the second center point according to the second formula, which is as follows: Where DM represents neighborhood density and VR represents the optimal denoising radius; DM reflects the degree of point cloud aggregation within a unit volume, distinguishing between "true sparsity" and "false sparsity", such as distant normal points and isolated noise points;

[0072] Step S304: Repeatedly obtain the number of neighborhood points and neighborhood density corresponding to all points in the third cubic region, and arrange them in ascending order, denoted as the neighborhood point count sequence and the neighborhood density sequence, respectively. Obtain the maximum value and median in the neighborhood point count sequence, and denot them as NM and NZ in order; obtain the median of the neighborhood density sequence, denoted as DZ; the median is not sensitive to outliers and can more reliably reflect the typical characteristics of the region, avoiding the influence of outliers on the mean; NM reflects the extreme density in the region and is used to set the upper limit of the threshold.

[0073] Step S305: Set the maximum number of points threshold to DS0 and the minimum number of points threshold to DS1; calculate the first number of points coefficient XE1, the second number of points coefficient XE2, and the third density coefficient XE3 respectively, where XE1=min(DS0, k5*NM), XE2=max(DS1, k6*NZ), XE3=1 / (k7*DZ); where k5, k6, and k7 are proportional coefficients; in this embodiment, k5=0.5, k6=0.1, and k7=2.5;

[0074] Step S306: Calculate the initial point threshold corresponding to the second center point according to the third formula, which is as follows: Where CY represents the initial point threshold; the initial point threshold for repeatedly acquiring all points within the third cubic region; high-density areas, such as walls, are allowed a higher point threshold, while low-density areas, such as vegetation, use a lower threshold;

[0075] Step S307: Calculate the average of the initial point count thresholds of all points excluding the second center point within the initial neighborhood of the second center point, denoted as the average of other thresholds CP; calculate the final point count threshold of the second center point according to the fourth formula, which is as follows: Where CZ represents the final point threshold, and q1 and q2 are weight coefficients; the final point threshold corresponding to all points in the first cube is repeatedly obtained; in this embodiment, q1=0.7, q2=0.3; the self-threshold reflects the local features of a single point, and the neighborhood mean reflects the global trend of the surrounding points. The fusion of the two can suppress the noise interference of a single point.

[0076] Step S308: For any point in the first cube, denote it as an arbitrary target point; denote the spherical region with the arbitrary target point as the center and the optimal denoising radius corresponding to the arbitrary target point as the radius as the arbitrary target neighborhood; if the number of points in the arbitrary target neighborhood is less than the final point count threshold corresponding to the arbitrary target point, then mark the arbitrary target point as a noise point; repeat the marking of all points in the first cube and remove the noise points to obtain the second point cloud data; for example, if the final point count threshold corresponding to a certain point is 5, and the number of points in the arbitrary target neighborhood of that point is 4, then mark that point as a noise point;

[0077] In practical implementation, XE1 is the maximum threshold requirement for high-density regions. That is, when the neighborhood density corresponding to the second center point is too high, the initial point number threshold output by the third formula is close to the upper limit, i.e., XE1 + XE2. XE1 can avoid setting too high a threshold in dense regions, which would cause valid points to be deleted, and limiting the upper limit of the threshold can reduce computational overhead. XE2 is the minimum protection level to ensure in extremely low-density regions. That is, when the neighborhood density corresponding to the second center point is close to 0, the minimum initial point number threshold output by the third formula is basically XE2. XE2 can ensure that basic retention is maintained in sparse point cloud regions during denoising, preventing excessive deletion and avoiding misjudging sparse valid points as noise. XE3 controls the sensitivity of the threshold to density changes, reflecting the threshold's response strength to density changes, and can be adjusted according to the point cloud density distribution characteristics.

[0078] Step S4 involves performing point cloud outlier detection processing based on the second point cloud data to obtain the third point cloud data; Step S4 includes the following sub-steps:

[0079] Step S401: Divide the first cube corresponding to the second point cloud data into k8 sub-cubes evenly, and denote any cube as the fourth cube region; denote any point in the fourth cube region as the first target point; where k8 is the set number of divisions; the fourth cube division ensures that the detection unit and the sub-region scale of the denoising stage are consistent or smaller, avoiding cross-region feature confusion;

[0080] Step S402: Obtain the g1 nearest points to the first target point within the fourth cubic region, denoted as the first neighboring points. Calculate the square of the Euclidean distance between each of the first neighboring points and the first target point, and sum them to obtain the sum of squared distances to the first target point, denoted as FS. In this embodiment, g1 is 10. The g1 nearest neighboring points focus on the direct proximity structure of the point, reflecting the compactness of the local spatial distribution. The sum of squared distances reflects the spatial dispersion of a point from its surrounding points within a local region; the larger the value, the more the point deviates from the local dense region.

[0081] Step S403: Repeatedly obtain the sum of squared distances of all points in the fourth cubic region and calculate the average to obtain the average of the sum of squared distances, denoted as FP;

[0082] Step S404: Calculate the squared distance ratio of the first target point according to the fifth formula, which is as follows: , where GS represents the distance squared ratio; the distance squared ratio of all points in the fourth cubic region is repeatedly obtained; GS eliminates the difference in FS caused by density differences in different sub-regions, and represents the degree of spatial deviation of the first target point relative to other points in the grid. The larger the value, the more "isolated" the point is in spatial distribution, and the more likely it is to be an outlier.

[0083] Step S405: Obtain the squared distance ratio of the first target point and its corresponding first neighboring point, denoted as the squared ratio sequence; calculate the mean and standard deviation of the squared ratio sequence, and denot them as GP and GB respectively in order.

[0084] Step S406: Calculate the dynamic threshold GY of the first target point based on GP and GB, GY = GP + g2 * GB, where g2 is a set proportional coefficient; if the squared distance ratio of the first target point is greater than the dynamic threshold GY of the first target point, then mark the first target point as an abnormal point; otherwise, mark the first target point as a normal point; in this embodiment, g2 is 2; the mean GP reflects the typical deviation of the sub-region, the standard deviation GB characterizes the deviation fluctuation range, and the dynamic threshold GY is automatically adjusted according to the local distribution. For example, in dense regions, GB is small and the threshold is compact; in sparse regions, GB is large and the threshold is loose; this avoids the defects of a globally fixed threshold and can adapt to areas with abrupt changes in point cloud density;

[0085] Step S407: Repeat the detection and marking of all points in the second point cloud data to obtain the third point cloud data.

[0086] In the specific implementation process, the first point cloud denoising process removes obvious noise, and the point cloud outlier detection process focuses on residual local outliers, such as small clusters of noise that were missed after denoising. The noise and outliers are gradually removed, avoiding the limitations of a single algorithm and improving the quality of point cloud data.

[0087] Example 2, please refer to Figure 4 As shown, Figure 4 A schematic diagram of an electronic device is provided, which may include a processor, a communication interface, a memory, and a communication bus. The processor, communication interface, and memory communicate with each other via the communication bus. The memory stores computer-readable instructions, and the processor can call these instructions. When the processor executes a computer-readable instruction, it performs steps such as those in a method for denoising and detecting outliers in UAV-borne LiDAR point cloud data, to achieve the following functions: collecting point cloud data of a target based on UAV-borne LiDAR and pre-dividing the point cloud data to obtain first point cloud data; performing radius acquisition processing based on the first point cloud data to obtain the corresponding optimal denoising radius, denoised as the denoising radius data; obtaining the corresponding final point count threshold based on the first point cloud data and performing first point cloud denoising processing based on the denoising radius data to obtain second point cloud data; and performing point cloud outlier detection processing based on the second point cloud data to obtain third point cloud data.

[0088] Furthermore, when the logical instructions in the aforementioned memory can be implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0089] Example 3: This application also provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it performs the steps of the above-described method for denoising and detecting outliers in UAV-borne LiDAR point cloud data to achieve the following functions: collecting point cloud data of a target based on UAV-borne LiDAR and pre-dividing the point cloud data to obtain first point cloud data; performing radius acquisition processing based on the first point cloud data to obtain the corresponding optimal denoising radius, denoised as denoising radius data; obtaining the corresponding final point count threshold based on the first point cloud data and performing first point cloud denoising processing based on the denoising radius data to obtain second point cloud data; and performing point cloud outlier detection processing based on the second point cloud data to obtain third point cloud data.

[0090] Based on the above description of the embodiments, the embodiments of the present invention can be provided as methods, systems, or computer program products. Based on this understanding, the above technical solutions, in essence or in terms of their contribution to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or certain parts of the embodiments.

[0091] In the embodiments provided in this application, it should be understood that the disclosed system or method can be implemented in other ways. The embodiments described above are merely illustrative. For example, the division of modules or units is only a logical functional division, and there may be other division methods in actual implementation. Furthermore, multiple modules or units may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the coupling or direct coupling or communication connection shown or discussed may be through some communication interfaces. The indirect coupling or communication connection between systems, modules, and units may be electrical, mechanical, or other forms.

[0092] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for denoising and outlier detection of UAV-borne lidar point cloud data, characterized in that, Includes the following steps: The first point cloud data is obtained by collecting point cloud data of the target using UAV-borne lidar and pre-dividing the point cloud data. Based on the first point cloud data, the radius is processed to obtain the corresponding optimal denoising radius, which is recorded as the denoising radius data. Based on the first point cloud data, the corresponding final point number threshold is obtained, and the first point cloud is denoised according to the denoising radius data to obtain the second point cloud data; The third point cloud data is obtained by performing point cloud outlier detection processing based on the second point cloud data. Based on the first point cloud data, the radius acquisition process is performed to obtain the corresponding optimal denoising radius, denoised as the denoising radius data. This includes the following sub-steps: Set the initial neighborhood radius to RE and the maximum neighborhood radius to RM; for any second cube, denote it as the second cubic region; for any point in the second cubic region, denote it as the first center point; The spherical region with the first center point as the center and RE as the radius is denoted as the initial neighborhood of the first center point; Obtain the distances from all points in the initial neighborhood of the first center point to the first center point, denoted as the neighborhood distance sequence; and calculate the average and variance of all distances in the neighborhood distance sequence, denoted as the corresponding mean PE and variance PB of the first center point in order. Repeatedly obtain the variance of all points within the second cubic region, and calculate the maximum value of the corresponding variance, denoted as PM; calculate the optimal denoising radius corresponding to the first center point according to the first formula; the first formula is as follows: Where RL represents the excellent noise reduction radius; Repeatedly obtain the excellent denoised radii corresponding to all points in the second cubic region, and arrange them in ascending order, denoted as the excellent radius sequence; calculate the average of the k2th percentile to the k3th percentile in the excellent radius sequence, and denoted as the suspected optimal radius corresponding to the second cubic region; where k2 and k3 are the set percentiles; If the suspected optimal radius corresponding to the second cubic region is less than the maximum neighborhood radius, then mark the suspected optimal radius corresponding to the second cubic region as the optimal denoising radius of the second cubic region; otherwise, mark the maximum neighborhood radius as the optimal denoising radius of the second cubic region. Repeatedly obtain the optimal denoising radius for all second cubes to get the denoising radius data.

2. The method for denoising and outlier detection of UAV-borne lidar point cloud data according to claim 1, characterized in that, The process of acquiring point cloud data of the target object based on UAV-borne LiDAR and pre-segmenting the point cloud data to obtain the first point cloud data includes the following sub-steps: For the target whose point cloud data is to be collected, it is denoted as the first target. The point cloud data of the first target is collected using the UAV's airborne lidar and is denoted as the raw point cloud data. Establish a three-dimensional spatial coordinate system and obtain the coordinates of all points in the original point cloud data in the three-dimensional spatial coordinate system, which are denoted as point cloud coordinate data.

3. The method for denoising and outlier detection of UAV-borne lidar point cloud data according to claim 2, characterized in that, The process of acquiring point cloud data of the target object based on UAV-borne LiDAR, and pre-segmenting the point cloud data to obtain the first point cloud data, also includes the following sub-steps: Obtain the extreme values ​​of the horizontal axis, vertical axis, and vertical axis coordinates in the point cloud coordinate data respectively, and construct the minimum bounding cube of the original point cloud data based on the extreme values ​​of the horizontal axis, vertical axis, and vertical axis coordinates, denoted as the first cube; The first cube is evenly divided into k1 sub-cubes, denoted as the second cube. After this is completed, the first point cloud data is obtained, where k1 is the set number of divisions.

4. The method for denoising and outlier detection of UAV-borne lidar point cloud data according to claim 3, characterized in that, The process of obtaining the final point count threshold based on the first point cloud data, and then performing denoising on the first point cloud based on the denoising radius data to obtain the second point cloud data includes the following sub-steps: The second cubic region is evenly divided into k4 sub-cubes, denoted as the third cube. Any third cube is denoted as the third cubic region, where k4 is the number of divisions set. For any point in the third cubic region, it is denoted as the second center point; the spherical region with the second center point as the center and the optimal noise reduction radius corresponding to the second center point as the radius is denoted as the initial neighborhood of the second center point. Count the number of points in the initial neighborhood of the second center point, denoted as the neighborhood number NA; and calculate the neighborhood density corresponding to the second center point according to the second formula, which is as follows: Wherein, DM represents neighborhood density and VR represents the optimal denoising radius; Repeatedly obtain the number of neighborhood points and the neighborhood density corresponding to all points in the third cubic region, and arrange them in ascending order, denoted as the neighborhood point number sequence and the neighborhood density sequence, respectively. Obtain the maximum value and median in the neighborhood point number sequence, and denot them as NM and NZ in order; obtain the median of the neighborhood density sequence, denoted as DZ.

5. The method for denoising and outlier detection of UAV-borne lidar point cloud data according to claim 4, characterized in that, The process of obtaining the final point count threshold based on the first point cloud data, and then performing denoising on the first point cloud based on the denoising radius data to obtain the second point cloud data, also includes the following sub-steps: Set the maximum number of points threshold to DS0 and the minimum number of points threshold to DS1; calculate the first number of points coefficient XE1, the second number of points coefficient XE2 and the third density coefficient XE3 respectively, where XE1 = min(DS0, k5*NM), XE2 = max(DS1, k6*NZ), XE3 = 1 / (k7*DZ); where k5, k6 and k7 are proportional coefficients; The initial point threshold corresponding to the second center point is calculated according to the third formula, which is as follows: CY=XE1*(1-e -(XE3*DM) )+XE2, where CY represents the initial point threshold; repeatedly obtain the initial point threshold for all points within the third cubic region; Calculate the average of the initial point count thresholds of all points in the initial neighborhood of the second center point, excluding the second center point, and denote it as the average of other thresholds CP; calculate the final point count threshold of the second center point according to the fourth formula, which is as follows: CZ=q1*CY+q2*CP, where CZ represents the final point count threshold, and q1 and q2 are weight coefficients; repeatedly obtain the final point count thresholds corresponding to all points in the first cube.

6. The method for denoising and outlier detection of UAV-borne lidar point cloud data according to claim 5, characterized in that, The process of obtaining the final point count threshold based on the first point cloud data, and then performing denoising on the first point cloud based on the denoising radius data to obtain the second point cloud data, also includes the following sub-steps: For any point in the first cube, denote it as an arbitrary target point; denote the spherical region with the arbitrary target point as the center and the optimal denoising radius corresponding to the arbitrary target point as the radius as the arbitrary target neighborhood. If the number of points in the arbitrary target neighborhood is less than the final number threshold corresponding to the arbitrary target point, then mark the arbitrary target point as a noise point; repeat the marking of all points in the first cube and remove the noise points to obtain the second point cloud data.

7. The method for denoising and outlier detection of UAV-borne lidar point cloud data according to claim 6, characterized in that, The process of detecting outliers in the second point cloud data to obtain the third point cloud data includes the following sub-steps: The first cube corresponding to the second point cloud data is evenly divided into k8 sub-cubes, and any one of the cubes is denoted as the fourth cube region; any point in the fourth cube region is denoted as the first target point; where k8 is the set number of divisions. Obtain the g1 points within the fourth cubic region that are closest to the first target point, and denote them as the first neighboring points. Calculate the square of the Euclidean distance between each of the first neighboring points and the first target point, and sum them to obtain the sum of squared distances to the first target point, denoted as FS. Repeatedly obtain the sum of squared distances of all points within the fourth cubic region, and calculate the average of the sums of squared distances. This average is denoted as FP. The ratio of the squared distance to the first target point is calculated using the fifth formula, which is as follows: Where GS represents the squared distance ratio; the squared distance ratio of all points in the fourth cubic region is repeatedly obtained.

8. The method for denoising and outlier detection of UAV-borne lidar point cloud data according to claim 7, characterized in that, The process of detecting outliers in the second point cloud data to obtain the third point cloud data also includes the following sub-steps: Obtain the squared ratio of the distance between the first target point and its corresponding first neighboring point, denoted as the squared ratio sequence. Calculate the mean and standard deviation of the squared ratio sequence, and denot them as GP and GB respectively in order. The dynamic threshold GY of the first target point is calculated based on GP and GB, where GY = GP + g2 * GB, and g2 is the set proportional coefficient. If the squared distance ratio of the first target point is greater than the dynamic threshold GY of the first target point, the first target point is marked as an abnormal point; otherwise, the first target point is marked as a normal point. Repeat the detection and labeling process for all points in the second point cloud data to obtain the third point cloud data.

Images