A filtering method for cast pipe surface three-dimensional point cloud data

Abstract

The application discloses a kind of filtering methods for the surface three-dimensional point cloud data of cast pipe, comprising the following steps;Step one, outlier search: find out the defect point contained in laser line point cloud data;Step two, least square straight line fitting, find out the reference of laser line data;Step three, distance set calculation: obtain the geometric characteristics of laser line point cloud data;The distance of each point on laser line to reference straight line is calculated, and the distance set is obtained;Step four, line laser point cloud data correction reference line generation: calculate each reference line on the reference surface of cylindrical development;Step five, point cloud data is projected to reference surface: obtain the corrected point cloud data;The obtained distance set is distributed with y i ′=a i ′x′ i +b i ′ as reference, obtain the cast pipe surface three-dimensional point cloud data of eliminating vibration noise and correcting tilt distortion.The application removes the tilt distortion and radial vibration noise in the original point cloud data of cast pipe surface, and provides high-quality point cloud data for subsequent cast pipe surface defect detection method.

Description

Technical Field

[0001] The present invention belongs to the technical field of denoising and distortion correction in point cloud and processing methods, and particularly relates to a filtering method for three-dimensional point cloud data of a cast pipe surface. Background Art

[0002] Centrifugal ductile iron pipes, a common piping material, are widely used in water conservancy projects, urban water supply, drainage, and natural gas transmission. To ensure product quality, hot cast pipes, fresh from the centrifuge, are inspected for surface defects, including cracks, pores, heavy skin, flow lines, and sink marks, during the production process. This allows quality issues to be identified early in the production process and unqualified products to be promptly removed from the production line, preventing them from flowing into the next process and wasting time and resources. This ensures product quality while reducing production costs.

[0003] After being produced using the centrifugal casting method, cast pipes are transported via a conveyor line to the next process. At this point, they are incandescent. Surface temperatures reach 500-700°C, making conventional contact inspection methods impossible. Furthermore, the small size of defects requires depth information to describe. For example, crack defects are approximately 3-100mm long, 0.1mm wide, and 0.2mm deep. Therefore, industrial camera-based visual inspection methods cannot capture defect images due to limited imaging accuracy.

[0004] When using a line laser sensor to acquire surface data on cylindrical workpieces such as cast pipes, the workpiece is typically rotated about its axis while the line laser sensor remains stationary, emitting a laser line parallel to the cylinder's axis. After the cylinder rotates one full revolution along its axis, the laser line captured by the line laser sensor is stitched together to form a complete rectangular image of the cylindrical surface.

[0005] 3D object data acquired by 3D line laser sensors is typically stored and processed in the form of point clouds. Point cloud data is defined as a set of vectors describing the geometric shape of an object in a 3D coordinate system. In addition to line laser sensors, point cloud data can also be acquired using devices such as 3D cameras and structured light scanners. These devices can intuitively represent the shape, surface color, and texture of objects in space with high precision, resolution, and dimensionality. Point cloud data is widely used in 3D modeling, scene reconstruction, robotic navigation, virtual reality, and augmented reality.

[0006] When acquiring point cloud data, the data collection system is subject to interference from environmental and human factors, which often results in the inclusion of abnormal noise and distorted data points. Point cloud filtering is the first step in point cloud data preprocessing, and its effectiveness significantly impacts subsequent processes.

[0007] The point cloud filtering algorithms proposed in the prior art are only designed for general situations and have good effects on general noise. For example, the grid smoothing algorithm can effectively remove surface noise from point cloud data of simple objects. However, for objects with complex surface shapes, in scenarios where multiple noise sources exist, the general three-dimensional point cloud filtering algorithm has a poor processing effect on point cloud noise. The reason is that the noise of point cloud data with different surface complexities and in different scenarios has its own special characteristics, and the general point cloud filtering algorithm does not take into account the characteristics of the collected object and the collection environment. The point cloud filtering algorithms given in the existing patents focus more on the actual characteristics of the object for design, but are not specifically designed for the three-dimensional point cloud data of the cast pipe surface with radial vibration and tilt distortion. As a result, the existing methods have poor filtering processing effects on the three-dimensional point cloud data of the cast pipe surface. Summary of the Invention

[0008] In order to overcome the problems existing in the above-mentioned prior art, the purpose of the present invention is to provide a filtering method for three-dimensional point cloud data of cast pipe surface, which provides high-quality point cloud data for subsequent cast pipe surface defect detection methods by removing tilt distortion and radial vibration noise in the original point cloud data of the cast pipe surface.

[0009] In order to achieve the above object, the technical solution adopted by the present invention is:

[0010] A filtering method for three-dimensional point cloud data of a cast pipe surface comprises the following steps:

[0011] Step 1: Outlier search: find the defect points contained in the point cloud data on the laser line;

[0012] Step 2: Least squares straight line fitting to find the benchmark of the laser line data;

[0013] Step 3: Distance set calculation: Get the geometric features of the laser line point cloud data; calculate each point p on the laser line i ∈l i To the reference line y i =a i x i +b i The distance is obtained by i ;

[0014] Step 4: Generate baselines for line laser point cloud data correction: Calculate each baseline on the cylindrical unfolded reference surface;

[0015] Step 5: Project the point cloud data onto the reference plane to obtain the corrected point cloud data; the distance set d obtained in step 3 is i y i ′=a i 'x' i +b i ' is used as the basis for distribution, and the three-dimensional point cloud data of the cast pipe surface with vibration noise eliminated and tilt distortion corrected is obtained. The distance features of the cast pipe surface point cloud data are projected onto a cylindrical surface expanded with the nominal diameter of the cast pipe.

[0016] The step 1 is specifically as follows:

[0017] For each collected laser line point cloud data set l i ={p i ,i∈[1,m]}, from the initial point p i Start neighborhood search and calculate point p i and point p i+1 The slope k of the straight line i , click p i+1 and point p i+2 The slope k of the straight line i+1 , calculate the difference q between the slopes of two adjacent lines i =k i -k i+1 ;

[0018] The acquisition steps are as follows: after the cast pipe enters the acquisition station, the in-place detection sensor starts the line laser sensor to collect a line of point cloud data on the cast pipe surface;

[0019] According to the scale characteristics of the convex surface of the cast pipe (due to the non-smooth cast pipe surface shape produced by the centrifugal casting process, the laser line point cloud data that does not contain the entire line presents a fluctuating shape), the threshold range of the slope difference between two adjacent points is determined to be [k min ,k max ]; if q i ∈[k min ,k max ], then p i+1 is a normal point; otherwise, the point is considered to be a defect point, marked as an outlier and recorded in an outlier set l oi In the meantime, record q i The indicator is used to judge the end of continuous outliers; then, continue with p i+1 Determine the next outlier point for the starting point; if the next outlier point p is found t+1 , then compare q t With q i Are the signs of the two the same? If they are different, then p i and pt+1 All points between are outliers and need to be marked; otherwise, only p is marked. t+1 As an outlier (an outlier is generated by a defect); until all points on the laser line are traversed, this step ends.

[0020] The step 2 is specifically as follows:

[0021] For laser line point cloud data i All points l that are not marked as outliers i \{p i ,i∈l oi} Perform least squares straight line fitting to obtain the value corresponding to the laser line l i The reference straight line equation y i =a i x+b i .

[0022] The step 4 is specifically as follows:

[0023] Given the formula for the surface of a cylinder (xa) 2 +(yb) 2 =r 2 ,z∈[0,h], the nominal value of the cylinder radius and height, according to the scale of the raised features on the cast tube surface, generate the reference straight line equation y for each line laser i ′=a i 'x' i +b i ′, and correct the distance change between the two laser lines caused by vibration.

[0024] The three-dimensional point cloud filtering method is used to filter the three-dimensional point cloud data of different types of workpieces with the same geometric shape. The line laser point cloud data tilt correction method is applied to the general situation where the line laser sensor tilts when collecting data; the vibration denoising method is used to process the three-dimensional point cloud data of any cylindrical rotating workpiece with radial vibration.

[0025] Beneficial effects of the present invention:

[0026] When performing tilt correction on laser line point cloud data, outliers caused by defects are found through neighborhood search. The interference of outliers is eliminated during straight line fitting, ensuring the accuracy of straight line fitting.

[0027] When denoising radial vibrations from laser line point cloud data, the nominal diameter of the cast pipe and the geometric features of the raised surfaces are used to unfold the cylindrical surface into multiple baselines. The line laser point cloud data is normalized, and the line laser distance set is distributed around the baselines, ensuring that the defect features of the cast pipe are not lost when eliminating radial vibrations.

[0028] The three-dimensional point cloud filtering method for cast pipe surfaces can, on the one hand, correct noise and distortion in the data, providing high-quality point cloud data for subsequent cast pipe surface defect detection, which helps improve the accuracy of subsequent algorithms. On the other hand, it compensates for hardware equipment installation errors through software algorithms, reduces the accuracy requirements for equipment installation, and improves the engineering application value of the entire point cloud data acquisition system. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 Schematic diagram of the three-dimensional point cloud acquisition device for the cast pipe surface.

[0030] Figure 2 Schematic diagram of the collected three-dimensional point cloud data of the cast pipe surface.

[0031] Figure 3 This figure shows the influence of the radial vibration of the casting tube and the tilt of the laser on the point cloud data.

[0032] Figure 4 This is the flow chart of the three-dimensional point cloud filtering algorithm for the cast pipe surface.

[0033] Figure 5 Schematic diagram of the outlier detection method based on neighborhood search.

[0034] Figure 6 Schematic diagram of the unfolded baseline for the cylindrical surface.

[0035] Figure 7 Schematic diagram of distance set mapping of laser line point cloud data.

[0036] Figure 8 The surface morphology of the cast pipe containing porosity defects.

[0037] Figure 9 This is a local point cloud data diagram of the cast pipe surface containing pore defects. (The pore defect data is circled by a black ellipse)

[0038] Figure 10 Results of outlier search results in the laser line. (A zoomed-in image shows outliers marked with red asterisks)

[0039] Figure 11 Comparison of the least squares straight line fitting results. (The left and right figures are the fitted lines without and with outliers, respectively.)

[0040] Figure 12 is the point cloud data after filtering. DETAILED DESCRIPTION

[0041] The present invention will be described in further detail below with reference to the accompanying drawings.

[0042] The structure of the cast pipe surface three-dimensional data acquisition system is as follows: Figure 1As shown, a casting tube (marked 3) rests horizontally on two rotating wheels (marked 4) on a turntable below, driven by the wheels to rotate around its own axis (marked 5). A line laser sensor (marked 1) is stationary above the casting tube, emitting a laser line (marked 2) that illuminates the casting tube's surface parallel to its axis of rotation. As the casting tube rotates, the laser line point cloud data collected by the line laser sensor is spliced ​​into a rectangular representation of the cylindrical surface.

[0043] like Figure 2 As shown in the figure. During the acquisition of point cloud data on the cast pipe surface, there are two main factors that affect the quality of the point cloud data: first, the mechanical error of the turntable in the acquisition system will cause radial vibration of the cast pipe during rotation; second, the deviation of the position of the line laser sensor will cause the collected line laser data to tilt. The above two situations make the collected point cloud data on the cast pipe surface contain noise and distortion, as shown in the figure. Figure 3 As shown in the figure, the defect data will be submerged by noise and distorted, resulting in the inability of subsequent defect detection algorithms to correctly detect cast pipe defects.

[0044] Therefore, the present application proposes a filtering method for three-dimensional point cloud data of the cast pipe surface to solve two problems in the process of cast pipe surface data acquisition. First, filter out the system noise of the Z-direction point cloud data caused by the vibration of the cast pipe; second, correct the Y-direction observation noise introduced by mechanical errors, including the robot movement position, laser installation and turntable tilt, which cause the laser line to be non-parallel to the cast pipe surface. Intuitively, under ideal circumstances, the three-dimensional point cloud data of the cast pipe surface collected by the system should be a rectangle expanded by a cylindrical surface in a horizontal direction. However, the vibration of the cast pipe will cause wrinkles in this rectangle, and the error in the laser acquisition posture will cause this rectangle to be non-parallel to the horizontal plane. The purpose of the three-dimensional point cloud filtering method given in this application is to smooth out the wrinkles of this rectangle and rotate the tilted rectangle to a horizontal position.

[0045] The process for collecting point cloud data on the cast pipe surface is as follows: First, after being cast from the centrifuge, the pipe is transported to the sand blowing station via a roller conveyor. A rotating mechanism below this station rotates the pipe along its axis. Second, after the in-position detection sensor outputs a signal, the robot drives the line laser sensor to a preset working distance above the pipe. As the pipe rotates, the line laser sensor is triggered to continuously collect 3D point cloud data from the pipe surface. Finally, the collected line point cloud data is spliced ​​together to obtain a complete 3D point cloud of the pipe surface, completing data collection.

[0046] Unlike the noise and distortion present in general point cloud data, the surface of the cast pipe is not a smooth surface, but rather has regular, tiny protrusions. This causes the collected point cloud data to contain surface defect data and normal data from the non-smooth surface to overlap. When performing laser line tilt correction, if the defect data that may be contained in the point cloud data is not excluded, the tilt correction will result in a large error. The radial vibration of the cast pipe is caused by the motion error of the mechanical system of the rotating station, making it impossible to obtain the statistical laws of vibration. Using uniform sampling to filter the vibration data will cause the point cloud data to undergo stretching or compression deformation, causing the defect morphology contained in the data to change.

[0047] The two characteristics of the above-mentioned cast pipe surface point cloud data lead to the fact that the general point cloud filtering method cannot guarantee that useful defect data in the point cloud data will not be lost during the filtering process. To address this problem, the three-dimensional point cloud data filtering method proposed in this application consists of five steps. The specific execution process is as follows: Figure 4 As shown. Assuming that there are m data points on each laser line of the line laser sensor, and the line laser sensor has collected n laser line point cloud data on the surface of the cast pipe, the collected point cloud data of the cast pipe surface is M×N points. The specific execution steps of this method are as follows:

[0048] Step 1: Outlier search: Find the defect points contained in the point cloud data of the laser line. i ={p i ,i∈[1,m]}, from the initial point p i Start neighborhood search and calculate point p i and point p i+1 The slope k of the straight line i , click p i+1 and point p i+2 The slope k of the straight line i+1 ,like Figure 5 Calculate the difference q between the slopes of two adjacent straight lines i =k i -k i+1 According to the scale characteristics of the convex surface of the cast pipe, the threshold range of the slope difference between two adjacent points is determined as [k min ,k max ]. If q i ∈[k min ,k max ], then p i+1 is a normal point. Otherwise, the point is considered a defect point, marked as an outlier and recorded in an outlier set l oi In the meantime, record q i The indicator is used to judge the end of continuous outliers. Then, continue with p i+1Determine the next outlier point for the starting point. If the next outlier point p is found t+1 , then compare q t With q i Are the signs of the two the same? If they are different, then p i and p t+1 All points between are outliers and need to be marked. Otherwise, only mark p t+1 As an outlier point. This step ends when all points on the laser line have been traversed;

[0049] Step 2, least squares straight line fitting: find the reference of laser line data. i All points l that are not marked as outliers i \{p i ,i∈l oi} Perform least squares straight line fitting to obtain the value corresponding to the laser line l i The reference straight line equation y i =a i x+b i ;

[0050] Step 3: Distance set calculation: Get the geometric features of the laser line point cloud data. Calculate each point p on the laser line i ∈l i To the reference line y i =a i x i +b i The distance is obtained by i ;

[0051] Step 4: Generate baselines for line laser point cloud data correction: Calculate each baseline on the cylindrical unfolded reference surface. Given the cylindrical surface formula (xa) 2 +(yb) 2 =r 2 ,z∈[0,h], the nominal value of the cylinder radius and height, according to the scale of the raised features on the cast tube surface, generate the reference straight line equation y for each line laser i ′=a i 'x' i +b i ′, and make corrections to the distance change between the two laser lines caused by vibration, such as Figure 6 As shown;

[0052] Step 5: Project the point cloud data onto the reference plane to obtain the corrected point cloud data. i y i ′=a i 'x' i +b i′ is used as the benchmark for distribution, and the three-dimensional point cloud data of the cast pipe surface with vibration noise eliminated and tilt distortion corrected is obtained, such as Figure 7 shown.

[0053] Finally, the rectangular 3D point cloud data of the filtered cast pipe surface is restored to the cylindrical cast pipe surface. It can be seen that after the filtering operation, the 3D point cloud data of the cast pipe surface retains the defect information contained in the original 3D point cloud data, providing high-quality point cloud data for subsequent defect detection algorithms.

[0054] The three-dimensional point cloud filtering method proposed in this application is not only effective for the three-dimensional point cloud data of the cast pipe surface, but can also be extended to the filtering of three-dimensional point cloud data of different types of workpieces with the same geometric shape. The line laser point cloud data tilt correction method in this application can be applied to the general situation where the line laser sensor is tilted when collecting data; the vibration denoising method in this application can be used to process the three-dimensional point cloud data of any cylindrical rotating workpiece with radial vibration.

[0055] Example:

[0056] The three-dimensional point cloud data filtering method proposed in the present invention is applied in a hot cast pipe surface defect detection system based on machine vision. By deploying a three-dimensional vision system based on a line laser sensor at the sand blowing station of a ductile iron centrifugal cast pipe production line, the three-dimensional point cloud data of the hot cast pipe surface is collected to detect whether there are defects on the cast pipe surface. The working process of the system is as follows: the pipes cast by high-speed centrifugal casting in a centrifugal ductile iron casting machine are transported to the sand blowing station through a conveyor line. An industrial robot is installed above the station, and a line laser sensor is installed at the end of the robot. The robot drives the line laser sensor to a suitable position above the cast pipe and then emits a laser line to illuminate the cast pipe surface. The two rotating wheels of the turntable below drive the cast pipe to rotate along its own axis, so that the line laser sensor can collect the complete three-dimensional point cloud data of the cast pipe surface.

[0057] In this cast pipe surface point cloud data acquisition system, the installation and motion accuracy of the mechanism directly impact the data collection quality. The data collected in the field environment contains two sources of error. First, the two rotating wheels of the lower turntable drive the cast pipe in axial rotation. Due to factors such as installation errors of the rotating mechanism, wear during use, and positional errors of the cast pipe itself, the cast pipe vibrates radially during axial rotation, causing defect data in the collected point cloud data to be distorted or obscured by vibration noise. Second, the robot-driven linear laser sensor, which remains at the acquisition position above the cast pipe, also introduces a certain degree of angular error. The laser line is not parallel to the pipe axis, causing the collected point cloud data to tilt, which poses difficulties for the cast pipe defect detection algorithm. These two factors combined lead to a decrease in the quality of the raw point cloud data, affecting the accuracy of the subsequent defect detection algorithm. Therefore, appropriate preprocessing methods are required to filter the collected point cloud data to eliminate interference during data acquisition. However, because the cast pipe surface is not smooth but rather exhibits regular, microscopic protrusions, conventional point cloud filtering algorithms are unable to effectively address these two sources of error in the existing point cloud data. Therefore, the point cloud filtering method proposed in this application is used to analyze the error sources of the acquisition system and the surface characteristics of the cast pipe, eliminate the two errors in the point cloud data, and at the same time keep the defect characteristics unaffected during the filtering process, thereby providing high-quality point cloud data for the subsequent cast pipe surface defect detection algorithm.

[0058] The surface morphology of the cast pipe is as follows Figure 8 As shown in the figure, the surface is not smooth due to the centrifugal casting process, but rather has regular small bumps. A set of point cloud data of the cast pipe surface containing pore defects was collected using the cast pipe surface defect detection system deployed on site. The local point cloud data containing pore defects was selected, as shown in the figure. Figure 9 The data contains 20 laser lines, each with 3200 points. The point cloud data of the porosity defect is circled in the figure.

[0059] The point cloud data filtering method processes the point cloud data. The processing process is as follows:

[0060] (1) Outlier search of laser line point cloud data: The threshold value of the slope difference for judging outliers is determined to be [-10, 10] based on the scale of the bulge on the cast pipe surface. Starting from the starting point of each laser line, the slope of the straight line formed by two adjacent points is calculated, the slope difference is calculated and compared with the slope difference threshold, and the data points greater than the threshold are marked as outliers. Continue to search for the next outlier and compare it with the slope of the first outlier. According to the slope sign between the two outliers, it is determined whether the two outliers are also outliers, until all 3200 data points on the laser line are traversed. Figure 10 As shown, outliers are marked in red;

[0061] (2) Least squares linear fitting of laser line point cloud data: The least squares method is used to fit the point cloud data excluding outliers. The slope of the fitted line is -0.01741 and the intercept is -126.1159. In addition, compared with the linear fitting results of the defect data, the slope of the fitted line containing defect data is -0.0174 and the intercept is -126.1249. Figure 11 As shown in the figure, the blue line on the left represents the straight line fitting result without outliers, while the green line on the right represents the straight line fitting result including outliers. There is a slight difference between the two fitting lines because the air hole defect data only occupies a small part of the entire laser line data. However, it confirms that the presence of defect data will cause deviations in the least squares straight line fitting results.

[0062] (3) Calculation of distance set of laser line point cloud data: traverse all points on the laser line point cloud data, calculate the distance from each point to the least squares fitting line, and obtain the point cloud data distance set;

[0063] (4) Generation of baselines for laser point cloud data: The collected object is a cast pipe with a diameter of 800 mm. Therefore, the surface of the cast pipe is unfolded according to the nominal diameter, and the expression of the baseline corresponding to the collected point cloud data is calculated in combination with the size of the protrusion on the cast pipe surface.

[0064] (5) Projecting the distance set of three-dimensional point cloud data onto the surface of the cast pipe: According to the baseline expression corresponding to each laser line, the data in the distance set is taken as the reference position with the baseline as the reference position, and the position coordinates of the point cloud data on each laser line are calculated.

[0065] After the above steps have traversed 20 laser lines, the three-dimensional point cloud data filtering of the cast pipe surface is completed. The filtered result is as follows: Figure 12 As shown, the black point cloud data represents the original, unfiltered local point cloud data of the cast pipe surface, while the blue point cloud data represents the filtered result. Comparing the two clearly shows that the filtered 3D point cloud data eliminates tilt and vibration noise while ensuring that the porosity defect data is unaffected by the filtering process. This filtered 3D point cloud data significantly reduces the complexity of subsequent defect detection algorithms and improves defect detection accuracy.

[0066] The present invention aims at the point cloud data tilt introduced by the equipment installation and moving parts position errors of the data acquisition system, including the robot, laser and cast pipe rotation mechanism, which causes the laser acquisition position to be non-parallel to the cast pipe rotation axis. A tilt correction method based on the surface characteristics of the cast pipe is designed to eliminate the tilt distortion of the three-dimensional point cloud data of the cast pipe surface. In the tilt correction method of the point cloud data on the cast pipe surface, a point cloud data outlier detection method based on neighborhood search is designed to find the defective data in the laser point cloud data of each line, exclude the defective point cloud data when performing the point cloud data tilt correction, and improve the accuracy of the tilt correction.

[0067] The present invention addresses the radial vibration noise of point cloud data caused by the motion error of the acquisition station rotation mechanism during data acquisition. Based on the geometric expansion relationship of the cast pipe surface and the nominal value of the cast pipe diameter, a radial vibration denoising method for point cloud data of the cast pipe surface is designed to eliminate the noise caused by radial vibration.

[0068] The present invention belongs to the denoising and distortion correction technology in point cloud and processing methods. In response to the needs of cast pipe surface defect detection, three-dimensional point cloud data of the cast pipe surface is collected. On the one hand, due to the low motion accuracy of the cast pipe rotation mechanism, the cast pipe vibrates when rotating along the axial direction, resulting in radial vibration noise contained in the point cloud data; on the other hand, due to the tilt of the acquisition posture of the line laser sensor, the laser line is not parallel to the axis of rotation of the cast pipe, resulting in the tilt of the collected point cloud data. In response to the above two problems, the present invention proposes a point cloud data filtering algorithm, which pre-processes the collected three-dimensional point cloud data of the cast pipe surface, effectively removes the radial vibration noise and tilt distortion in the point cloud data, and provides high-quality data support for subsequent defect detection algorithms. In response to the noise in the point cloud data collection of the cast pipe surface, the present invention combines the geometric characteristics of the cast pipe surface morphology and the characteristics of the point cloud collection equipment to design a point cloud data filtering method suitable for the cast pipe surface. The cylindrical surface point cloud data collected when the three-dimensional point cloud data of the cast pipe surface has radial vibration and the line laser sensor acquisition position is tilted is filtered to remove the point cloud data deformation caused by the above factors, providing high-quality point cloud data for the subsequent cast pipe surface defect detection algorithm, thereby improving the algorithm's execution efficiency and detection accuracy.

Claims

1. A filtering method for three-dimensional point cloud data of a cast pipe surface, characterized in that: The following steps are included: Step 1: Outlier search: find the defect points contained in the point cloud data on the laser line; Step 2: For laser line point cloud data i Perform a least squares straight line fit on all points that are not marked as outliers to find the benchmark of the laser line data; Step 3: Distance set calculation: obtain the geometric features of the laser line point cloud data; Calculate for each point p on the laser line i ∈l i To the reference line y i =a i x i +b i The distance is obtained by i ; Step 4: Generate baselines for line laser point cloud data correction: Calculate each baseline on the cylindrical unfolded reference surface; Given the formula for the surface of a cylinder (xa) 2 +(yb) 2 =r 2 ,z∈[0,h], r is the nominal value of the cylinder radius, h is the nominal value of the height, and the reference straight line equation y for each line laser is generated according to the scale of the raised features on the cast pipe surface i ′=a i 'x' i +b i ′, and correct the distance change between the two laser lines caused by vibration; Step 5: Project the point cloud data onto the reference plane to obtain the corrected point cloud data; The distance set d obtained in step 3 is i y i ′=a i 'x' i +b i ′ is used as the benchmark for distribution, and the three-dimensional point cloud data of the cast pipe surface with vibration noise eliminated and tilt distortion corrected is obtained.

2. The filtering method for three-dimensional point cloud data of a cast pipe surface according to claim 1, characterized in that: The step 1 is specifically as follows: For each collected laser line point cloud data set l i ={p i ,i∈[1,m]}, from the initial point p i Start neighborhood search and calculate point p i and point p i+1 The slope k of the straight line i , click p i+1 and point p i+2 The slope k of the straight line i+1 , calculate the difference q between the slopes of two adjacent lines i =k i -k i+1 ; The acquisition steps are as follows: after the cast pipe enters the acquisition station, the in-place detection sensor triggers the line laser sensor to collect a line of point cloud data on the cast pipe surface; According to the scale characteristics of the convex surface of the cast pipe, the threshold range of the slope difference between two adjacent points is determined as [k min ,k max ]; if q i ∈[k min ,k max ], then it is believed that p i+1 is a normal point; otherwise, the point is considered to be a defect point, marked as an outlier and recorded in an outlier set l oi In the meantime, record q i The indicator is used to judge the end of continuous outliers; then, continue with p i+1 Determine the next outlier point for the starting point; if the next outlier point p is found t+1 , then compare q t With q i Are the signs of the two the same? If they are different, then p i and p t+1 All points in between are outliers and need to be marked; Otherwise, just mark p t+1 as an outlier point; until all points on the laser line are traversed, the step ends.

3. The filtering method for three-dimensional point cloud data of a cast pipe surface according to claim 1, characterized in that: The step 2 is specifically as follows: For laser line point cloud data i All points l that are not marked as outliers i ={p i ,i∈l oi } Perform least squares straight line fitting to obtain the value corresponding to the laser line l i The equation of the reference line y i =a i x+b i .

4. A filtering method for three-dimensional point cloud data of a cast pipe surface according to any one of claims 1 to 3, characterized in that: The three-dimensional point cloud data filtering method is used to filter the three-dimensional point cloud data of different types of workpieces with the same geometric shape. The line laser point cloud data tilt correction method is applied to the general situation where the line laser sensor tilts when collecting data; the vibration denoising method is used to process the three-dimensional point cloud data of any cylindrical rotating workpiece with radial vibration.

Images