Noise reduction processing method for three-dimensional point cloud data

Abstract

The invention discloses a noise reduction processing method for three-dimensional point cloud data. The method comprises the steps that S1, an initial point cloud picture of a stock bin and a first point cloud picture after the material amount in the stock bin changes are acquired; s2, performing threshold segmentation on the initial point cloud image and the first point cloud image to obtain a second point cloud image and a third point cloud image; s3, denoising the second point cloud atlas and the third point cloud atlas through a Z-axis denoising method, S4, performing statistical filtering on the fourth point cloud atlas and the fifth point cloud atlas through statistical filtering based on a KD tree nearest neighbor search method to obtain a sixth point cloud atlas and a seventh point cloud atlas; and S5, combining the sixth point cloud picture and the seventh point cloud picture into a closed eighth point cloud picture, wherein the eighth point cloud picture is used for calculating the variable quantity of the material quantity in the material bin. Noisy points in the point cloud data are effectively removed, and then the volume can be accurately calculated.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a noise reduction processing method for three-dimensional point cloud data. Background Art

[0002] In the field of animal husbandry, the feed stored in the silo is often corn and wheat. These feeds will generate huge dust during storage, making it difficult to accurately and stably monitor the feed level. Or the feed level is too high, causing safety accidents such as overflow.

[0003] During each feeding process, a lot of dust will be generated, which will seriously affect the correct value of the radar scanning and produce some discrete points. These discrete deviation points make it impossible for the laser radar to correctly scan the three-dimensional spatial shape of the object, and thus cannot accurately reconstruct the shape of the inside of the discharge tower. Accurately constructing the shape of the inside of the discharge tower is the basis for establishing a 3D model of the material surface and calculating the material volume.

[0004] The laser radar level meter is a non-contact continuous measurement that can accurately scan the three-dimensional shape of an object. Because some silos are too large and the dust is too heavy, the laser radar penetration is not enough, and there will be errors in the scan value and noise. At the same time, there are one or more cables inside the material tower for measuring temperature and humidity. These cables will shake with the feeding and discharging of materials. These noise points scanned out seriously affect the accuracy of the subsequent calculation of the material level.

[0005] Denoising is the most basic and important part of point cloud data preprocessing. Due to the defects of the radar scanning equipment itself, scanning errors, and foreign matter in the silo itself, noise points will inevitably appear during the scanning process, and these error values ​​may seriously affect the validity of the three-dimensional point cloud data. Based on this, the present invention proposes a denoising processing method for three-dimensional point cloud data. Summary of the invention

[0006] The present invention provides a denoising method for three-dimensional point cloud data, which effectively removes noise points in point cloud data without affecting the three-dimensional point cloud shape of the object to be measured. It can process different noise points of point cloud images scanned inside different material towers or inside the same material tower, form a correct and complete three-dimensional point cloud shape, and then accurately calculate the volume.

[0007] According to one aspect of the present disclosure, a method for denoising three-dimensional point cloud data is provided, the method comprising: S1, obtaining an initial point cloud image of the silo and a first point cloud image after the amount of material in the silo changes; S2, performing threshold segmentation on the initial point cloud image and the first point cloud image to obtain a second point cloud image and a third point cloud image; S3, denoising the second point cloud image and the third point cloud image by a Z-axis denoising method, including: performing Z-axis calibration on the second point cloud image and the third point cloud image to obtain a point cloud image perpendicular to the Z-axis, and deleting the messy discrete points at the top of the point cloud image according to the Z-axis coordinate size to obtain a fourth point cloud image and a fifth point cloud image; S4, performing statistical filtering on the fourth point cloud image and the fifth point cloud image by statistical filtering based on a KD tree nearest neighbor search method to obtain a sixth point cloud image and a seventh point cloud image; S5, merging the sixth point cloud map and the seventh point cloud map into a closed eighth point cloud map, and the eighth point cloud map is used to calculate the change in the amount of material in the silo.

[0008] In a possible implementation, the step S2, performing threshold segmentation on the initial point cloud image and the first point cloud image to obtain a second point cloud image and a third point cloud image, includes: Calculate the shortest distance between each point in the initial point cloud and the first point cloud to obtain the shortest distance array D1; Create a Boolean mask M1, set a threshold T1, select the points in array D1 whose distance is greater than the threshold T1, and obtain the second point cloud image; Calculate the shortest distance between each point in the first point cloud and the initial point cloud to obtain the shortest distance array D1; Create a Boolean mask M2, set a threshold T2, select the points in array D2 whose midpoint distance is greater than the threshold, and obtain the third point cloud image.

[0009] In a possible implementation, calculating the shortest distance from each point in the initial point cloud image to the first point cloud image to obtain the shortest distance array D1 includes: For the initial point cloud And the first point cloud , for each point ,turn up Make the Euclidean distance d minimum, the specific formula is as follows: (1) in Yes and The Euclidean distance between them is as follows: (2) In formula (2), x, y, and z represent three-dimensional coordinates, and i and j represent the serial numbers of points in different point clouds. The distance array is obtained by formula (2): ,in Yes The nearest point The distance between.

[0010] In a possible implementation, creating a Boolean mask M1, setting a threshold T1, selecting points in the array D1 whose midpoint distance is greater than the threshold T1, and obtaining a second point cloud image includes: Create a Boolean mask M1 as shown in formula (3), set a threshold T1 as shown in formula (4), and select the points in array D1 whose distance is greater than the threshold; (3) (4) (5) Among them, the coefficient Used to adjust the threshold value according to the actual situation. represents the mean of the distances in array D1; According to the mask M1, the points corresponding to the initial point cloud P are filtered out to obtain the second point cloud P new2 : (6).

[0011] In a possible implementation, the second point cloud image P is obtained by using new2 The same method is used to obtain the third point cloud image P new3 .

[0012] In a possible implementation, S3 performs Z-axis calibration on the second point cloud image and the third point cloud image to obtain a point cloud image perpendicular to the Z axis, and deletes the messy discrete points at the top of the point cloud image according to the Z-axis coordinate size to obtain a fourth point cloud image and a fifth point cloud image, including: Suppose the coordinate matrix of the points in the point cloud is P, whose dimension is , where N is the number of points and each row represents the three-dimensional coordinates (x, y, z) of a point; (7) Get the roll angle value roll from the attitude sensor installed on the radar and calculate the rotation matrix. The rotation matrix R is: (8) Use the rotation matrix R to rotate the point cloud matrix P, that is: (9) in is the transposed matrix of the rotation matrix R, Represents the rotated point cloud coordinate matrix, whose dimension is ; After Z-axis calibration, a point cloud image perpendicular to the Z-axis is obtained, and the fourth point cloud image is obtained by deleting the messy discrete points at the top according to the size of the Z-axis; the fifth point cloud image is obtained by the same method as that of the fourth point cloud image.

[0013] In a possible implementation, S4, performing statistical filtering on the fourth point cloud image and the fifth point cloud image by statistical filtering based on a nearest neighbor search method to obtain a sixth point cloud image and a seventh point cloud image, including: Use KD tree to find the point in the fourth point cloud The K nearest neighbor points of ; (10) Wherein, N is the total number of points in the fourth point cloud image; For each point Calculate it to the k nearest neighbors The average Euclidean distance of is calculated as follows: (11) in, Yes To the nearest point The Euclidean distance of The formula for the statistical filtering threshold is: (12) In the formula, std is the standard deviation coefficient, which is used to control the influence of the standard deviation on the distance threshold. For all The average value of is: (13) In the formula is the standard deviation of the distance: (14) The points whose distance is greater than the filtering threshold are removed to obtain the sixth point cloud image. The fifth point cloud image is processed in the same way to obtain the seventh point cloud image.

[0014] Compared with the prior art, the present invention has the following beneficial effects: A denoising method for three-dimensional point cloud data in an embodiment of the present disclosure can process noise points of point cloud images scanned inside different material towers or the same material tower to form a correct and complete three-dimensional point cloud shape, and then accurately calculate the volume. Accurately constructing the shape of the inside of the discharge tower is the basis for establishing a 3D model of the material surface and calculating the volume of the material. It is used in the field of animal husbandry to monitor the feed level in the silo, and can also be used to calculate the material levels of various solids and liquids in the industrial field.

[0015] Through multi-stage point cloud processing, combined with machine learning algorithms, the accuracy and applicability of denoising can be further enhanced. It can denoise point cloud images very well. It is not only suitable for point cloud images of silos, but also for point cloud scans in the fields of stockpiling, chemical industry, etc., effectively solving the problem of noise interference. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A flowchart of a method for reducing noise of three-dimensional point cloud data according to an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0017] Various exemplary embodiments, features and aspects of the present disclosure will be described in detail below with reference to the accompanying drawings. The same reference numerals in the accompanying drawings represent elements with the same or similar functions. Although various aspects of the embodiments are shown in the accompanying drawings, the drawings are not necessarily drawn to scale unless otherwise specified.

[0018] The word “exemplary” is used exclusively herein to mean “serving as an example, example, or illustration.” Any embodiment described herein as “exemplary” is not necessarily to be construed as preferred or advantageous over other embodiments.

[0019] In addition, in order to better illustrate the present disclosure, numerous specific details are given in the following specific embodiments. It should be understood by those skilled in the art that the present disclosure can also be implemented without certain specific details. In some examples, methods, means, components and circuits well known to those skilled in the art are not described in detail in order to highlight the subject matter of the present disclosure.

[0020] According to one aspect of the present disclosure, a method for denoising three-dimensional point cloud data is provided, the method comprising: S1, obtaining an initial point cloud image of the silo and a first point cloud image after the amount of material in the silo changes; S2, performing threshold segmentation on the initial point cloud image and the first point cloud image to obtain a second point cloud image and a third point cloud image; S3, denoising the second point cloud image and the third point cloud image by a Z-axis denoising method, including: performing Z-axis calibration on the second point cloud image and the third point cloud image to obtain a point cloud image perpendicular to the Z-axis, and deleting the messy discrete points at the top of the point cloud image according to the Z-axis coordinate size to obtain a fourth point cloud image and a fifth point cloud image; Each coordinate of the three-dimensional point cloud data has three values ​​(x, y, z), and the z-axis coordinate size is the value of the coordinate z in the vertical direction.

[0021] S4, performing statistical filtering on the fourth point cloud image and the fifth point cloud image by statistical filtering based on a KD tree nearest neighbor search method to obtain a sixth point cloud image and a seventh point cloud image; S5, merging the sixth point cloud map and the seventh point cloud map into a closed eighth point cloud map, and the eighth point cloud map is used to calculate the change in the amount of material in the silo.

[0022] In a possible implementation, the step S2, performing threshold segmentation on the initial point cloud image and the first point cloud image to obtain a second point cloud image and a third point cloud image, includes: Calculate the shortest distance between each point in the initial point cloud and the first point cloud to obtain the shortest distance array D1; Create a Boolean mask M1, set a threshold T1, select the points in array D1 whose distance is greater than the threshold T1, and obtain the second point cloud image; Calculate the shortest distance between each point in the first point cloud and the initial point cloud to obtain the shortest distance array D1; Create a Boolean mask M2, set a threshold T2, select the points in array D2 whose midpoint distance is greater than the threshold, and obtain the third point cloud image.

[0023] In a possible implementation, calculating the shortest distance from each point in the initial point cloud image to the first point cloud image to obtain the shortest distance array D1 includes: For the initial point cloud And the first point cloud , for each point ,turn up Make the Euclidean distance d minimum, the specific formula is as follows: (1) in Yes and The Euclidean distance between them is as follows: (2) In formula (2), x, y, and z represent three-dimensional coordinates, and i and j represent the serial numbers of points in different point clouds. The distance array is obtained by formula (2): ,in Yes The nearest point The distance between.

[0024] In a possible implementation, creating a Boolean mask M1, setting a threshold T1, selecting points in the array D1 whose midpoint distance is greater than the threshold T1, and obtaining a second point cloud image includes: Create a Boolean mask M1 as shown in formula (3), set a threshold T1 as shown in formula (4), and select the points in array D1 whose distance is greater than the threshold; (3) (4) (5) Among them, the coefficient Used to adjust the threshold value according to the actual situation. represents the mean of the distances in array D1; If the threshold is too large, effective points will be lost; if the threshold is too small, the noise reduction effect will not be obvious.

[0025] According to the mask M1, the points corresponding to the initial point cloud P are filtered out to obtain the second point cloud P new2 : (6).

[0026] In a possible implementation, the second point cloud image P is obtained by using new2 The same method is used to obtain the third point cloud image P new3 .

[0027] Threshold segmentation can effectively remove the same parts in the two point cloud images and obtain the different parts of the point cloud images caused by the change of material quantity in the silo.

[0028] In the actual point cloud scanning of the silo, during the feeding process, there are often some dense and far-distance point clusters that are not removed at the top due to excessive dust. Therefore, for all point clouds obtained in step S3, the Z-axis denoising method is further used. The part of the point cloud that is close to the zero point is directly deleted. The useless part of the scanned point cloud is further removed.

[0029] In a possible implementation, S3 performs Z-axis calibration on the second point cloud image and the third point cloud image to obtain a point cloud image perpendicular to the Z axis, and deletes the messy discrete points at the top of the point cloud image according to the Z-axis coordinate size to obtain a fourth point cloud image and a fifth point cloud image, including: Suppose the coordinate matrix of the points in the point cloud is P, whose dimension is , where N is the number of points, and each row represents the three-dimensional coordinates (x, y, z) of a point; (7) Get the roll angle value roll from the attitude sensor installed on the radar and calculate the rotation matrix. The rotation matrix R is: (8) Use the rotation matrix R to rotate the point cloud matrix P, that is: (9) in is the transposed matrix of the rotation matrix R, Represents the rotated point cloud coordinate matrix, whose dimension is ; After Z-axis calibration, a point cloud image perpendicular to the Z-axis is obtained, and the fourth point cloud image is obtained by deleting the messy discrete points at the top according to the size of the Z-axis; the fifth point cloud image is obtained by the same method as that of the fourth point cloud image.

[0030] The Z-axis calibration method is to calculate the rotation matrix of the points of the point cloud by obtaining the value of the roll angle roll transmitted by the attitude sensor installed on the radar, so that the scanned point cloud image is perpendicular to the ground.

[0031] The Z-axis denoising method removes the noise points at the top of the point cloud, and there will be some discrete points that are not within the range of the Z-axis. The statistical filtering is used in combination with the machine learning algorithm to remove discrete points. This method is based on the average distance from a point to all neighboring points within a certain range, which will not exceed the threshold and filter out outliers that do not meet the requirements. It has better processing effects when the noise points are unevenly distributed and the density difference is relatively large.

[0032] In a possible implementation, S4, performing statistical filtering on the fourth point cloud image and the fifth point cloud image by statistical filtering based on a nearest neighbor search method to obtain a sixth point cloud image and a seventh point cloud image, including: Use KD tree to find the point in the fourth point cloud The K nearest neighbor points of : (10) Where N is the total number of points in the fourth point cloud image; For each point Calculate it to the k nearest neighbors The average Euclidean distance of is calculated as follows: (11) in, Yes To the nearest point The Euclidean distance of The formula for the statistical filtering threshold is: (12) In the formula, std is the standard deviation coefficient, which is used to control the influence of the standard deviation on the distance threshold. For all The average value of is: (13) In the formula is the standard deviation of the distance: (14) The points whose distance is greater than the filtering threshold are removed to obtain the sixth point cloud image. The fifth point cloud image is processed in the same way to obtain the seventh point cloud image.

[0033] The nearest neighbor search used in the statistical filtering method uses an exhaustive method to search for the nearest point, and its time complexity is O(N). Therefore, the KD tree is used to speed up the nearest neighbor search, and a point can be found in O(logN) time. The nearest neighbor can significantly reduce the complexity compared to the O(N) direct brute force search.

[0034] The KDTree method is further used to quickly search the neighborhood, and then the points that are too far away are removed by setting a threshold, thus obtaining a more accurate point cloud map.

[0035] Through multi-stage point cloud processing, combined with machine learning algorithms, the accuracy and applicability of denoising can be further enhanced. It can denoise point cloud images very well. It is not only suitable for point cloud images of silos, but also for point cloud scans in the fields of stockpiling, chemical industry, etc., effectively solving the problem of noise interference.

[0036] The flow chart and block diagram in the accompanying drawings show the possible architecture, function and operation of the system, method and computer program product according to multiple embodiments of the present disclosure. In this regard, each square box in the flow chart or block diagram can represent a part of a module, program segment or instruction, and a part of the module, program segment or instruction includes one or more executable instructions for realizing the specified logical function. In some alternative implementations, the function marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two continuous square boxes can actually be executed substantially in parallel, and they can sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs the specified function or action, or can be implemented with a combination of special hardware and computer instructions.

[0037] The embodiments of the present disclosure have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and changes will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The selection of terms used herein is intended to best explain the principles of the embodiments, practical applications, or technical improvements in the market, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein.

Claims

1. A method for denoising three-dimensional point cloud data, characterized in that: The method comprises: S1, obtaining an initial point cloud image of the silo and a first point cloud image after the amount of material in the silo changes; S2, performing threshold segmentation on the initial point cloud image and the first point cloud image to obtain a second point cloud image and a third point cloud image; S3, denoising the second point cloud image and the third point cloud image by a Z-axis denoising method, including: performing Z-axis calibration on the second point cloud image and the third point cloud image to obtain a point cloud image perpendicular to the Z-axis, and deleting the messy discrete points at the top of the point cloud image according to the Z-axis coordinate size to obtain a fourth point cloud image and a fifth point cloud image; S4, performing statistical filtering on the fourth point cloud image and the fifth point cloud image by statistical filtering based on a KD tree nearest neighbor search method to obtain a sixth point cloud image and a seventh point cloud image; S5, merging the sixth point cloud map and the seventh point cloud map into a closed eighth point cloud map, and the eighth point cloud map is used to calculate the change in the amount of material in the silo.

2. A method for denoising three-dimensional point cloud data according to claim 1, characterized in that: The step S2, performing threshold segmentation on the initial point cloud image and the first point cloud image to obtain a second point cloud image and a third point cloud image, comprises: Calculate the shortest distance between each point in the initial point cloud and the first point cloud to obtain the shortest distance array D1; Create a Boolean mask M1, set a threshold T1, select the points in array D1 whose distance is greater than the threshold T1, and obtain the second point cloud image; Calculate the shortest distance between each point in the first point cloud and the initial point cloud to obtain the shortest distance array D1; Create a Boolean mask M2, set a threshold T2, select the points in array D2 whose midpoint distance is greater than the threshold, and obtain the third point cloud image.

3. The method for denoising three-dimensional point cloud data according to claim 2, characterized in that: The calculating the shortest distance from each point in the initial point cloud image to the first point cloud image to obtain the shortest distance array D1 includes: For the initial point cloud And the first point cloud , for each point ,turn up Make the Euclidean distance d minimum, the specific formula is as follows: （1） in Yes and The Euclidean distance between them is as follows: （2） In formula (2), x, y, and z represent three-dimensional coordinates, and i and j represent the serial numbers of points in different point clouds. The distance array is obtained by formula (2): ,in Yes The nearest point The distance x between them.

4. The method for denoising three-dimensional point cloud data according to claim 2, characterized in that: The method of creating a Boolean mask M1, setting a threshold T1, selecting points in the array D1 whose midpoint distance is greater than the threshold T1, and obtaining a second point cloud image includes: Create a Boolean mask M1 as shown in formula (3), set a threshold T1 as shown in formula (4), and select the points in array D1 whose distance is greater than the threshold; （3） （4） （5） Among them, the coefficient Used to adjust the threshold value according to the actual situation. represents the mean of the distances in array D1; According to the mask M1, the points corresponding to the initial point cloud P are filtered out to obtain the second point cloud P new2 : （6）。 5. The method for denoising three-dimensional point cloud data according to claim 4, characterized in that: Use to get the second point cloud P new2 The same method is used to obtain the third point cloud image P new3 .

6. The method for denoising three-dimensional point cloud data according to claim 1, characterized in that: In S3, the Z-axis vertical point cloud images are obtained by performing Z-axis calibration on the second point cloud image and the third point cloud image, and the messy discrete points at the top of the point cloud image are deleted according to the Z-axis coordinate size to obtain the fourth point cloud image and the fifth point cloud image, including: Suppose the coordinate matrix of the points in the point cloud is P, whose dimension is , where N is the number of points and each row represents the three-dimensional coordinates (x, y, z) of a point; （7） Get the roll angle value roll from the attitude sensor installed on the radar and calculate the rotation matrix. The rotation matrix R is: （8） Use the rotation matrix R to rotate the point cloud matrix P, that is: (9) in is the transposed matrix of the rotation matrix R, Represents the rotated point cloud coordinate matrix, whose dimension is ; After Z-axis calibration, a point cloud image perpendicular to the Z-axis is obtained, and the fourth point cloud image is obtained by deleting the messy discrete points at the top according to the size of the Z-axis; the fifth point cloud image is obtained by the same method as that of the fourth point cloud image.

7. The method for denoising three-dimensional point cloud data according to claim 1, characterized in that: S4, performing statistical filtering on the fourth point cloud image and the fifth point cloud image by statistical filtering based on a nearest neighbor search method to obtain a sixth point cloud image and a seventh point cloud image, including: Use KD tree to find the point in the fourth point cloud The K nearest neighbor points of : （10） Where N is the total number of points in the fourth point cloud image; For each point Calculate it to the k nearest neighbors The average Euclidean distance of is calculated as follows: （11） in, Yes To the nearest point The Euclidean distance of The formula for the statistical filtering threshold is: (12) In the formula, std is the standard deviation coefficient, which is used to control the influence of the standard deviation on the distance threshold. For all The average value of is: (13) In the formula is the standard deviation of the distance: （14） The points whose distance is greater than the filtering threshold are removed to obtain the sixth point cloud image. The fifth point cloud image is processed in the same way to obtain the seventh point cloud image.

Images