CN117934429A - Flatness detection method of bus side wall panel based on 3D point cloud and contour matching - Google Patents

Abstract

The invention discloses a passenger car side wall board flatness detection method based on three-dimensional point cloud and contour matching, which comprises the following steps: acquiring one-frame three-dimensional point clouds of side wallboards of different types of passenger cars; preprocessing the three-dimensional point cloud to generate point cloud slice data; performing line segment fitting on the point cloud slice data, and storing a line segment function obtained by fitting, the number of acquired points and region division into a template library; performing point cloud preprocessing operation when collecting each frame of point cloud data of a measured piece, matching the point cloud preprocessing operation with point cloud slice data in a template library, setting a relevant defect threshold value, and judging whether a point is a defect point or not; unfolding the data into complete point cloud data of the side wall plate of the passenger car, and dividing out a defect area; and eliminating misjudgment points by using an abnormal point eliminating algorithm to obtain a defect area. The invention solves the problems that the detection time is too long due to the oversized side wall plate of the passenger car, and the detection is difficult due to the too small defects by adopting the contour matching method, thereby meeting the real-time and accurate defect detection on the complex curved surface of the side wall plate of the passenger car.

Description

Flatness detection method of bus side wall panel based on 3D point cloud and contour matching

Technical Field

The invention belongs to the field of machine vision nondestructive testing, and in particular is a method for detecting the flatness of a passenger car side wall panel based on three-dimensional point cloud and contour matching.

Background technique

In the manufacturing process of large parts, manufacturing defects are inevitable, such as texture damage on the surface of the object, welds produced during the welding process of materials, and dents caused by extrusion and collision. If these defects are not discovered in time and maintained and repaired, the service life of the product will be greatly reduced, and even bring great safety risks. Taking the side wall panels of passenger cars as an example, the various defects on the side wall panels will greatly affect the strength and appearance of the side wall panels. Large defects may even cause the risk of paint and putty falling off during the operation of the train. The appearance of the car body is mostly white. The use of machine vision defect detection technology is not effective for detecting small defects with depth information, while point cloud defect detection technology can well meet the requirements for extracting defect features of side wall panels.

Product surface quality inspection based on point cloud is a field worthy of in-depth research, and it is also a key step in manufacturing and product use and maintenance. 3D point cloud defect detection technology can be summarized into five categories: 1. Based on contour point cloud; 2. Based on template matching; 3. Based on local geometric features; 4. Based on multimodal point cloud data; 5. Based on deep learning. These point cloud defect detection methods have their own advantages: based on contour point cloud, it has strong adaptability and fast speed; based on template matching, it is accurate and suitable for standard parts, but the speed is slow; based on local geometric features, it has high generalization ability; multimodal point cloud has high accuracy and strong generalization; based on deep learning, it can process irregular point clouds, but it is difficult to obtain samples and has low versatility. The side wall panels of buses are large in size, the defect features are small and require high precision. If the above methods are used, it is difficult to detect small defects and the detection time is too long.

Summary of the invention

The technical problem to be solved by the present invention is to provide a bus side wall panel flatness detection method based on three-dimensional point cloud and contour matching in response to the above-mentioned deficiencies in the prior art. The method slices the point cloud data and processes the sliced point cloud data, thereby solving the problem of long detection time caused by the large size of the bus side wall panels. At the same time, the contour matching method is adopted to solve the problem of difficult detection due to small defects. It can meet the needs of real-time and accurate defect detection on complex curved surfaces such as bus side wall panels, and provide reliable data support for the subsequent putty spraying on the surface of the bus side wall panels.

In order to achieve the above technical objectives, the technical solution adopted by the present invention is:

A method for detecting the flatness of a passenger car side wall panel based on three-dimensional point cloud and contour matching comprises the following steps:

Step 1, obtaining a frame of 3D point cloud data of different types of standard passenger car side wall panels;

Step 2: pre-process the acquired 3D point cloud data, and use the Y coordinate as the main direction to generate a point cloud slice data;

Step 3: Fit the point cloud slice data to obtain two straight line segments and two curved line segments, record the fitted straight line segment function, curved line segment function, the number of points collected and the area division, and store them in the template library;

Step 4: perform point cloud preprocessing when collecting each frame of point cloud data of the tested object. Take the Y coordinate of each frame of point cloud data as the reference, extract the corresponding point cloud slice data from the template library made in step 3 and match it with the first frame of point cloud data collected from the tested object. Calculate the shortest distance from each point in each frame of point cloud to the line segment corresponding to the matching bus type in the template library, set the relevant defect threshold, and when the calculated shortest distance is greater than the relevant defect threshold, the point is identified as a defect point, and the position and parameters of the point are recorded;

Step 5: After all the point cloud slice data are collected, they are expanded into complete bus side wall panel point cloud data according to the Y coordinate, and the defective area is segmented;

Step 6: Use the outlier elimination algorithm to eliminate misjudged points and finally obtain the defect area information.

As a further improved technical solution of the present invention, the step 1 is specifically:

The bus side wall panel scanning device is used to obtain a frame of three-dimensional point cloud data of different types of standard bus side wall panels.

As a further improved technical solution of the present invention, the step 2 is specifically as follows:

Step 2.1, filtering and denoising a frame of standard point cloud data obtained in step 1, removing abnormal points and noise in the side wall panel point cloud, and downsampling the point cloud data using a uniform downsampling method;

Step 2.2: Adjust the pose of the side wall panel point cloud data after preprocessing, taking the Y coordinate axis as its main direction, that is, all points in the point cloud data have the same Y coordinate.

As a further improved technical solution of the present invention, the step 3 is specifically as follows:

Step 3.1, taking the point cloud slice data obtained in step 2 as the object, assuming that there are k points in the slice data P, since the Y coordinate values of each point are the same, the points in the slice are regarded as a point set in two-dimensional data, that is, each point consists of an x coordinate and a z coordinate, then P = {(x ₁ ,z ₁ ),(x ₂ ,z ₂ ),…,(x _k ,z _k )};

Step 3.2, divide the set P into four parts according to the x-coordinate, namely _P1 , _P2 , _P3 , _P4 , where _P1 and _P3 are curve point sets, _P2 and _P4 are straight line point sets, assuming that there are m points in _P1 , n points in _P3 , a points in _P2 , b points in _P4 , and m+n+a+b=k;

Step 3.3: Use the least squares method to fit the straight line part in the slice data:

l(x)＝a ₀ +a ₁ x (1);

Where l(x) is the fitted straight line function. Each point in the data to be fitted is substituted into the least squares formula (1) to calculate the predicted value. The weighted square error is then used to determine whether the straight line fitting is reasonable. The calculation formula is as follows:

Where ω( _xi )>0 is the weight function within the range _{of xi} , l( _xi ) is the fitted function value, f( _xi ) is the original value, and _xi is the coordinate of the i-th point; when the weighted error sum of the fitted straight line segment is When is minimum, the least square fitting is completed;

Use formula (1) and formula (2) to fit P ₂ , and fit the straight line segment function l ₁ (x). Use formula (1) and formula (2) to fit P ₄ , and fit the straight line segment function l ₂ (x).

Step 3.4: Use the circle fitting method to obtain the curve segment equation in the slice data:

(Xx) ² +(Zz) ² =r ² (3);

The coordinates of the center of the circle are (x, z) and the radius of the circle is r. The fitting process is divided into the following steps:

Step 3.4.1, first randomly select three different points in the curve segment corresponding to the curve part point set, and use the principle of three-point circle to fit the curve segment circle equation;

Step 3.4.2: Since the curve segment in the slice data is less than a quarter of a circle, the fitted curve segment circle equation is transformed into a function z=f(x), and the points of the curve segment in the slice data are brought into the fitted function to obtain a new z coordinate value, and at the same time, the determination coefficient R ² is obtained:

Among them, _zi represents the true z coordinate value of the i-th point, Indicates the predicted z coordinate value of the i-th point, /> Represents the average value of the true z coordinates of all points;

When R ² <0.95, return to step 3.4.1 and randomly select three points for curve fitting again until R ² ≥0.95, and the curve fitting is completed;

Use steps 3.4.1 and 3.4.2 to fit P ₁ and obtain the curve segment function f ₁ (x). Use steps 3.3.1 and 3.3.2 to fit P ₃ and obtain the curve segment function f ₂ (x).

Step 3.5, repeat steps 3.1 to 3.4 to obtain the two straight line segment functions, two curve segment functions, the number of points in the point cloud slice data and the area division of all bus types and their side wall panel point cloud slice data, and store the fitted two straight line segment functions, two curve segment functions, the number of points in the point cloud slice data and the area division and the corresponding bus types into the template library.

As a further improved technical solution of the present invention, the step 4 is specifically as follows:

Step 4.1, using the bus side wall panel scanning device to collect the three-dimensional point cloud data of the measured part, when the first frame of point cloud data is collected, immediately perform the preprocessing method in step 2.1 on it, and at the same time use the point cloud posture adjustment method in step 2.2 to make the Y coordinates of each point in the same frame of point cloud data the same;

Step 4.2, calculate the number of points in the first frame of point cloud slice data collected, find the bus type that is closest to the number of points in the first frame of point cloud slice data calculated in the template library, and match it;

Step 4.3, extract two straight line segment functions and two curve segment functions corresponding to the matching bus type from the template library, and divide the collected point cloud slice data of the tested object according to the areas in the template library, and divide the set P corresponding to the point cloud slice data of the tested object into four parts _P1 , _P2 , _P3 , and _P4 according to the x-coordinate, where _P1 and _P3 are curve part point sets, and _P2 and _P4 are straight line part point sets, and calculate the shortest distance from each point in each frame of the acquired point cloud slice data of the tested object to the corresponding line segment function in the matching bus type;

Step 4.4: When calculating the shortest distance from a point in the straight line portion of the measured object to the straight line segment function, assuming that the straight line segment function l ₁ (x) is transformed into a straight line equation ax+bz+c=0, the shortest distance from the point to the straight line segment is:

In formula (5), distance _i is the shortest distance from point (x _i ,z _i ) to line segment l ₁ (x);

Using the method in step 4.4, find the shortest distance from each point in the _P2 point set of the tested object to the straight line segment function _l1 (x) corresponding to the matching bus type in the template library, and find the shortest distance from each point in the _P4 point set of the tested object to the straight line segment function _l2 (x) corresponding to the matching bus type in the template library;

Step 4.5: When finding the shortest distance from a point in the curve part point set of the measured object to the curve segment function, it is necessary to find the perpendicular point q ₀ between the point and the curve. Assume that at a point P _i ( _xi , z _i ) in the P ₁ point set of the measured object, select any point q _τ on the curve segment function f ₁ (x), and find the direction vector of the tangent line at point q _τ. The direction vector of the line segment formed by P _i (x _i ,z _i ) and q _τ is/> Set the relevant threshold k, if/> Then q _τ is the perpendicular point q ₀ between the desired point _Pi ( _xi , _z1 ) and the curve segment function _f1 (x). Otherwise, add a step size to search for the next point on the curve segment function _f1 (x), and repeat the steps of finding the perpendicular point until it satisfies/> After obtaining the perpendicular point q ₀ =(x,z), the shortest distance from point _Pi ( _xi , _zi ) to the curve segment function _f1 (x) can be obtained:

In formula (6), distance _i is the shortest distance from point P _i (x _i ,z _i ) to the curve segment function f ₁ (x);

Using the method in step 4.5, find the shortest distance from each point in the _P1 point set of the tested object to the curve segment function _f1 (x) corresponding to the bus type matched in the template library, and find the shortest distance from each point in the _P3 point set of the tested object to the curve segment function _f2 (x) corresponding to the bus type matched in the template library;

Step 4.6, set the relevant defect threshold ε, and compare the shortest distance from each point in the _P2 point set of the tested piece to the straight line segment function _l1 (x) corresponding to the bus type matched in the template library, the shortest distance from each point in the _P4 point set of the tested piece to the straight line segment function _l2 (x) corresponding to the bus type matched in the template library, the shortest distance from each point in the _P1 point set of the tested piece to the curve segment function _f1 (x) corresponding to the bus type matched in the template library, and the shortest distance from each point in the _P3 point set of the tested piece to the curve segment function _f2 (x) corresponding to the bus type matched in the template library with the relevant defect threshold ε. If a shortest distance is greater than the relevant defect threshold ε, the point in the tested piece corresponding to the shortest distance is a defect point, and the shortest distance corresponding to the defect point and the position of the defect point are recorded;

Step 4.7: After completing steps 4.1 and 4.2, each time a frame of point cloud slice data of the test piece is collected, steps 4.2 to 4.6 are cycled once.

As a further improved technical solution of the present invention, the step 5 is specifically as follows:

Step 5.1, after collecting all the point cloud slice data of the side wall panel of the test piece, all the collected point cloud slice data are expanded in the Y direction to obtain the point cloud data of a whole side wall panel, and the defect area composed of defect points is marked.

As a further improved technical solution of the present invention, the step 6 is specifically as follows:

Step 6.1, extracting all defect areas, and dividing the extracted defect areas into defect area 1, defect area 2, ... defect area n;

Step 6.2, assuming that the set of points in the defect area j is Q, j = 1, 2, 3 ... n, calculate the minimum bounding box of all points in the set Q, obtain the volume T of the minimum bounding sphere, set a relevant threshold γ, when T < γ, the defect area is considered to be a misjudged area and corrected to a non-defective area;

Step 6.3: After all defective areas are detected according to step 6.2, the misjudged areas are eliminated to finally obtain the defective areas.

The beneficial effects of the present invention are:

(1) The present invention establishes a template library for bus side wall panels and stores different types of standard bus side wall panels in the template library, so that the algorithm can adapt to the flatness detection of different types of bus side wall panels. Even if a brand new bus side wall panel is manufactured, it only needs to be scanned once and stored in the template library for detection, thus solving the problem of poor adaptability.

(2) The present invention processes the outline of each frame of point cloud instead of processing the entire point cloud after scanning, thereby avoiding the problem of taking too long to process the entire point cloud data. This makes it possible to extract defective areas while collecting the side wall panel point cloud data, greatly reducing the required running time.

(3) The present invention uses point cloud contours for template matching, which amplifies the defect features in the detection and solves the problem of low defect detection accuracy and inability to detect small defects due to the large size of the bus side wall panels.

(4) The present invention splices all the obtained point cloud slice data into complete side wall panel point cloud data, marks the defective area, and then uses the abnormal area elimination algorithm to correct the misjudged defective area into a non-defective area, thereby solving the problem of errors caused by hardware reasons. At the same time, the coordinates and depth information of the defective area can be obtained, providing a reliable data source for the subsequent side wall panel putty surface spraying.

(5) The present invention slices the point cloud data and processes the sliced point cloud data, thereby solving the problem of long detection time caused by the large size of the bus side wall panel. At the same time, the contour matching method is adopted to solve the problem that defects are difficult to detect due to being too small. It can meet the needs of real-time and accurate defect detection on complex curved surfaces such as bus side wall panels, and provide reliable data support for the subsequent putty spraying on the surface of the bus side wall panels.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 is an overall flow chart of a method for detecting flatness of a side wall panel of a passenger car according to the present invention.

Figure 2 is a contour diagram of the standard point cloud of the side wall panel.

Figure 3 is a point cloud contour diagram of the side wall panel defects.

Figure 4 is the complete point cloud of the side wall panel

FIG5 is a diagram of the actual defect area in FIG4 .

Figure 5 a is an enlarged view of the actual defect area in Figure 4.

Figure 5 (b) is a pure real defect area map.

Detailed ways

The specific embodiments of the present invention are further described below according to the accompanying drawings:

A method for detecting the flatness of a bus side wall panel based on three-dimensional point cloud and contour matching, as shown in FIG1 , comprises the following steps:

Step 1, obtaining a frame of 3D point cloud data of different types of standard passenger car side wall panels;

Step 2: pre-process the acquired 3D point cloud data, and use the Y coordinate as the main direction to generate a point cloud slice data;

Step 3: Fit the point cloud slice data to obtain two straight line segments and two curved line segments, record the fitted straight line segment function, curved line segment function, the number of points collected and the area division, and store them in the template library;

Step 4: perform point cloud preprocessing when collecting each frame of point cloud data of the tested object. Take the Y coordinate of each frame of point cloud data as the reference, extract the corresponding point cloud slice data from the template library made in step 3 and match it with the first frame of point cloud data of the tested object. Calculate the shortest distance from each point in each frame of point cloud of the tested object to the line segment corresponding to the matching bus type in the template library, set the relevant defect threshold, and when the calculated shortest distance is greater than the relevant defect threshold, the point is identified as a defect point, and the position and parameters of the point are recorded;

Step 5: After all the point cloud slice data are collected, they are expanded into complete bus side wall panel point cloud data according to the Y coordinate, and the defective area is segmented;

Step 6: Use the outlier elimination algorithm to eliminate misjudged points and finally obtain the defect area information.

The step 1 is specifically as follows:

Step 1.1, build a passenger car side wall panel scanning device (using the existing structure), including scanning equipment (camera), base, bracket, cantilever beam, servo motor, etc. The base and bracket structure of the scanning equipment are mainly composed of independent slide rail 1, independent slide rail 2, cantilever beam connector, support frame body and base, etc. Independent slide rail 1 and independent slide rail 2 are used to control the distance between the camera and the object to be measured; considering the resolution and effective acquisition area of the line laser scanning camera, a total of six camera placement points are set on the cantilever beam camera support frame to ensure that the entire side wall panel surface data can be collected at one time;

Step 1.2: There are four servo motors in total to control the scanning movement of the entire device. Two of the motors are placed at the junction of the support frame and the cantilever beam to control the forward and backward movement and up and down movement of the cantilever beam and adjust the distance between the line laser scanning camera and the workpiece under test. The remaining two motors are used to control the movement of the base on the slide rail, thereby realizing the movement of the line laser scanning camera to collect three-dimensional data of the workpiece under test.

The step 2 is specifically as follows:

Step 2.1, perform filtering and denoising preprocessing operations on a frame of standard point cloud data obtained in step 1, remove abnormal points and noise in the side wall panel point cloud, and use a uniform downsampling method to downsample the point cloud data to reduce subsequent processing time;

Step 2.2, adjust the pose of the side wall panel point cloud data after preprocessing, taking the Y coordinate axis as its main direction, that is, the Y coordinates of all points in the point cloud data are the same;

The step 3 is specifically as follows:

Step 3.1, taking the point cloud slice data obtained in step 2 as the object, assuming that there are k points in the slice data P, since the Y coordinate values of each point are the same, the points in the slice are regarded as a point set in two-dimensional data, that is, each point consists of an x coordinate and a z coordinate, then P = {(x ₁ ,z ₁ ),(x ₂ ,z ₂ ),…,(x _k ,z _k )};

Step 3.2, as shown in Figure 2, which is a contour diagram of the standard point cloud of the side wall panel, the set P is divided into four parts _P1 , _P2 , _P3 , and _P4 according to the x-coordinate, where _P1 and _P3 are curve part point sets, and _P2 and _P4 are straight line part point sets. Assume that there are m points in _P1 , n points in _P3 , a points in _P2 , and b points in _P4 , and m+n+a+b=k;

Step 3.3: Use the least squares method to fit the straight line part in the slice data:

l(x)＝a ₀ +a ₁ x (1);

Where l(x) is the fitted straight line function. According to general experience, l(x) is a linear function. Each point in the data to be fitted is substituted into the least squares formula (1) to calculate the predicted value. The weighted error sum of squares is then used to determine whether the straight line fitting is reasonable. The calculation formula is as follows:

Where ω( _xi )>0 is the weight function within the range _{of xi} , l( _xi ) is the fitted function value, f( _xi ) is the original value, and _xi is the coordinate of the i-th point; when the weighted error sum of the fitted straight line segment is When is minimum, the least square fitting is completed;

Use formula (1) and formula (2) to fit P ₂ , and fit the straight line segment function l ₁ (x). Use formula (1) and formula (2) to fit P ₄ , and fit the straight line segment function l ₂ (x).

Step 3.4: It can be seen from the manufacturing process that the curve part in the contour of the bus side wall panel is similar to an arc, so the circle fitting method will be used to obtain the curve segment equation in the slice data:

(Xx) ² +(Zz) ² =r ² (3);

The coordinates of the center of the circle are (x, z) and the radius of the circle is r. The fitting process is divided into the following steps:

Step 3.4.1, first randomly select three different points in the curve segment corresponding to the curve part point set, and use the principle of three-point circle to fit the curve segment circle equation;

Step 3.4.2: Since the curve segment in the slice data is obviously smaller than a quarter of a circle, the fitted curve segment circle equation can be transformed into a function z=f(x), where one x corresponds to one z. Substitute the points of the curve segment in the slice data into the fitted function to obtain the new z coordinate value and the determination coefficient R ² :

Among them, _zi represents the true z coordinate value of the i-th point, Represents the predicted z coordinate value of the ith point, that is, the z value obtained by the fitted function, /> Represents the average value of the true z coordinates of all points;

When R ² <0.95, return to step 3.4.1 and randomly select three points for curve fitting again until R ² ≥0.95, and the curve fitting is completed;

Use steps 3.4.1 and 3.4.2 to fit P ₁ and obtain the curve segment function f ₁ (x). Use steps 3.3.1 and 3.3.2 to fit P ₃ and obtain the curve segment function f ₂ (x).

Step 3.5, repeat steps 3.1 to 3.4 to obtain the two straight line segment functions, two curve segment functions, the number of points in the point cloud slice data and the area division of all bus types and their side wall panel point cloud slice data, and store the fitted two straight line segment functions, two curve segment functions, the number of points in the point cloud slice data and the area division and the corresponding bus types into the template library.

The step 4 is specifically as follows:

Step 4.1, using the bus side wall scanning device, collect the three-dimensional point cloud data of the measured object along the ground track. When the first frame of point cloud data is collected, immediately perform the preprocessing method in step 2.1 on it, and at the same time use the point cloud posture adjustment method in step 2.2 to make the Y coordinates of each point in the same frame of point cloud data the same;

Step 4.2: According to experience, the number of points in the point cloud slice data of different types of bus side wall panels varies greatly. Therefore, the number of points in the first frame of point cloud slice data collected is calculated, and the bus type with the closest number of points to the calculated first frame of point cloud slice data is found in the template library and matched;

Step 4.3, extract two straight line segment functions and two curve segment functions corresponding to the matching bus type from the template library, and divide the collected point cloud slice data of the tested object according to the areas in the template library, and divide the set P corresponding to the point cloud slice data of the tested object into four parts _P1 , _P2 , _P3 , and _P4 according to the x-coordinate, where _P1 and _P3 are curve part point sets, and _P2 and _P4 are straight line part point sets, and calculate the shortest distance from each point in each frame of the acquired point cloud slice data of the tested object to the corresponding line segment function corresponding to the matching bus type;

As shown in Figure 3, the set corresponding to the point cloud slice data of the tested part is divided into four parts P ₁ , P ₂ , P ₃ , and P ₄ according to the x coordinate. Figure 3 is a point cloud contour diagram of the side wall panel defect;

Step 4.4: When calculating the shortest distance from a point in the straight line portion of the measured object to the straight line segment function, assuming that the straight line segment function l ₁ (x) is transformed into a straight line equation ax+bz+c=0, the shortest distance from the point to the straight line segment is:

In formula (5), distance _i is the shortest distance from point (x _i ,z _i ) to line segment l ₁ (x);

Using the method in step 4.4, find the shortest distance from each point in the _P2 point set of the tested object to the straight line segment function _l1 (x) corresponding to the matching bus type in the template library, and find the shortest distance from each point in the _P4 point set of the tested object to the straight line segment function _l2 (x) corresponding to the matching bus type in the template library;

Step 4.5: When finding the shortest distance from a point in the curve part point set of the measured object to the curve segment function, it is necessary to find the perpendicular point q ₀ between the point and the curve. Assume that at a point P _i ( _xi , z _i ) in the P ₁ point set of the measured object, select any point q _τ on the curve segment function f ₁ (x), q _τ represents the τth point on the curve segment function f ₁ (x), and find the direction vector of the tangent at point q _τ The direction vector of the line segment formed by P _i (x _i ,z _i ) and q _τ is/> Set the relevant threshold k, if/> Then q _τ is the perpendicular point q ₀ between the desired point _Pi ( _xi , _z1 ) and the curve segment function _f1 (x). Otherwise, add a step size to search for the next point on the curve segment function _f1 (x). Repeat the steps of finding the perpendicular point until/> is satisfied. After obtaining the perpendicular point q ₀ =(x,z), the shortest distance from point _Pi ( _xi , _zi ) to the curve segment function _f1 (x) can be obtained:

In formula (6), distance _i is the shortest distance from point P _i (x _i ,z _i ) to the curve segment function f ₁ (x);

Using the method in step 4.5, find the shortest distance from each point in the _P1 point set of the tested object to the curve segment function _f1 (x) corresponding to the bus type matched in the template library, and find the shortest distance from each point in the _P3 point set of the tested object to the curve segment function _f2 (x) corresponding to the bus type matched in the template library;

Step 4.6, set the relevant defect threshold ε, and compare the shortest distance from each point in the _P2 point set of the tested piece to the straight line segment function _l1 (x) corresponding to the bus type matched in the template library, the shortest distance from each point in the _P4 point set of the tested piece to the straight line segment function _l2 (x) corresponding to the bus type matched in the template library, the shortest distance from each point in the _P1 point set of the tested piece to the curve segment function _f1 (x) corresponding to the bus type matched in the template library, and the shortest distance from each point in the _P3 point set of the tested piece to the curve segment function _f2 (x) corresponding to the bus type matched in the template library with the relevant defect threshold ε. If a shortest distance is greater than the relevant defect threshold ε, the point in the tested piece corresponding to the shortest distance is a defect point, and the shortest distance corresponding to the defect point and the position (parameters) of the defect point are recorded;

Step 4.7: After completing steps 4.1 and 4.2, each time a frame of point cloud slice data of the test piece is collected, steps 4.2 to 4.6 are cycled once.

The step 5 is specifically as follows:

Step 5.1, when the bus side wall panel scanning device is completed, that is, when all the point cloud slice data of the side wall panel of the tested part is collected, all the collected point cloud slice data are expanded in the Y direction to obtain a whole piece of side wall panel point cloud data, and the defect area composed of defect points is marked.

The step 6 is specifically as follows:

Step 6.1, due to the occasional camera shake, noise not completely eliminated and other reasons, misjudgment will inevitably occur, so all defect areas are extracted, and the extracted defect areas are divided into defect area 1, defect area 2, ... defect area n;

Step 6.2, assuming that the set of points in the defect area j is Q, j = 1, 2, 3 ... n, calculate the minimum bounding box of all points in the set Q, obtain the volume T of the minimum bounding sphere, set a relevant threshold γ, when T < γ, the defect area is determined to be a misjudged defect area, and it is corrected to a non-defective area; as shown in Figure 4, including the misjudged defect area and the real defect area; a in Figure 5 is an enlarged view of the real defect area in Figure 4, and b in Figure 5 is the extracted pure real defect area;

Step 6.3: After all defect areas are detected according to step 6.2, the misjudged areas are eliminated, and finally the defect areas and defect parameters are obtained.

The present invention establishes a coach side wall panel template library and stores different types of coach side wall panels in the template library, so that the method can adaptively detect the flatness of different types of coach side wall panels. Even if a brand new coach side wall panel is manufactured, it only needs to be scanned once and stored in the template library for detection, thus solving the problem of poor adaptability.

The present invention processes the outline of each frame of point cloud instead of processing the entire point cloud after scanning, thereby avoiding the problem of taking too long to process the entire point cloud data, making it possible to extract defective areas while collecting the side wall panel point cloud data, greatly reducing the required running time.

The present invention uses point cloud contours for template matching, amplifies defect features in detection, and solves the problem of low defect detection accuracy and inability to detect small defects due to the excessive size of the bus side wall panel.

The present invention splices all the obtained point cloud slice data into complete side wall panel point cloud data, marks the defective area, and then uses the abnormal area elimination algorithm to correct the misjudged defective area into a non-defective area, thereby solving the problem of errors caused by hardware reasons. At the same time, the coordinates and depth information of the defective area can be obtained, providing a reliable data source for subsequent side wall panel putty surface spraying.

The protection scope of the present invention includes but is not limited to the above embodiments. The protection scope of the present invention shall be based on the claims. Any replacement, deformation, and improvement of the technology that can be easily thought of by technicians in this field shall fall within the protection scope of the present invention.

Claims (7)

1. A passenger car side wall board flatness detection method based on three-dimensional point cloud and contour matching is characterized by comprising the following steps:

step 1, acquiring one-frame three-dimensional point cloud data of side wallboards of different types of standard buses;

Step 2, preprocessing the obtained three-dimensional point cloud data, and generating a part of point cloud slice data by taking a Y coordinate as a main direction;

Step 3, carrying out line segment fitting on the point cloud slice data, fitting two straight line segments and two curve segments, recording the straight line segment function and the curve segment function obtained by fitting, and the number and the area division of the acquired points, and storing the obtained points and the area division into a template library;

Step 4, performing point cloud preprocessing operation when collecting each frame of point cloud data of the measured piece, extracting corresponding point cloud slice data from the template library which is made in the step 3 by taking the Y coordinate as a reference, matching the corresponding point cloud slice data with the collected first frame of point cloud data of the measured piece, calculating the shortest distance from each point in each frame of point cloud to a line segment which corresponds to the matched bus type in the template library, setting a relevant defect threshold, and when the calculated shortest distance is larger than the relevant defect threshold, determining the point as a defect point and recording the position and parameters of the point;

Step 5, after all point cloud slice data are acquired, expanding the point cloud slice data into complete point cloud data of the side wall plate of the passenger car according to Y coordinates, and dividing a defect area in the point cloud data;

and 6, eliminating misjudgment points by using an abnormal point eliminating algorithm, and finally obtaining the defect area information.

2. The method for detecting the flatness of the side wall panel of the passenger car based on the three-dimensional point cloud and the contour matching according to claim 1, wherein the step 1 is specifically as follows:

and taking one-frame three-dimensional point cloud data of the side wallboards of different types of standard buses by using the scanning device of the side wallboards of the buses.

3. The method for detecting the flatness of the side wall panel of the passenger car based on the three-dimensional point cloud and the contour matching according to claim 1, wherein the step2 is specifically as follows:

step 2.1, filtering and denoising the frame of standard point cloud data obtained in the step 1, removing abnormal points and noise in the point cloud of the side wall plate, and performing downsampling on the point cloud data by using a uniform downsampling method;

and 2.2, adjusting the pose of the point cloud data of the side wall panel after the pretreatment is finished, and taking the Y coordinate axis as the main direction of the pose, namely, the Y coordinates of all points in the point cloud data are the same.

4. The passenger car side wall panel flatness detection method based on three-dimensional point cloud and contour matching according to claim 1, wherein the step 3 specifically comprises:

step 3.1, taking the point cloud slice data obtained in the step 2 as an object, assuming that k points exist in the slice data P, and considering the points in the slice as a point set in the two-dimensional data because the Y coordinate values of the points are the same, i.e., each point is composed of an x coordinate and a z coordinate, then p= { (x ₁,z₁),(x₂,z₂),...,(x_k,z_k) };

Step 3.2, dividing the set P into four parts P ₁、P₂、P₃、P₄ according to the x coordinate, wherein P ₁ and P ₃ are curve part point sets, P ₂ and P ₄ are straight part point sets, assuming that there are m points in P ₁, n points in P ₃, a point in P ₂, b points in P ₄, and m+n+a+b=k;

step 3.3, fitting the straight line part in the slice data by using a least square method:

l(x)＝a₀+a₁x (1)；

Wherein l (x) is a fitted linear function, each point in the data to be fitted is brought into a least square method formula (1) to calculate a predicted value, and whether the linear fitting is reasonable or not is judged by a weighted error square sum The calculation formula of (2) is as follows:

Wherein ω (x _i) > 0 is the weight function in the range of x _i, l (x _i) is the fitted function value, f (x _i) is the original value, and x _i is the coordinate of the i-th point; when the weighted error square sum of the fitted straight line segments And if the least square is the least, the least square fitting is completed;

fitting P ₂ by using a formula (1) and a formula (2) to obtain a straight-line segment function l ₁ (x), and fitting P ₄ by using the formula (1) and the formula (2) to obtain a straight-line segment function l ₂ (x);

and 3.4, obtaining a curve segment equation in slice data by adopting a circle fitting method:

(X-x)²+(Z-z)²＝r² (3)；

Wherein the center coordinates are (x, z), and the radius of the circle is r; the fitting process comprises the following steps:

Step 3.4.1, firstly randomly picking out different three points in a curve segment corresponding to a curve part point set, and fitting out a curve segment circle equation by utilizing the principle of three-point circle setting;

In step 3.4.2, since the curve segment in the slice data is smaller than one quarter circle, the fitted curve segment circle equation is converted into a function z=f (x), the points of the curve segment in the slice data are brought into the fitted function, a new z coordinate value is obtained, and a determination coefficient R ² is obtained at the same time:

Wherein z _i represents the true z-coordinate value of the ith point, Predictive z-coordinate value representing the ith point,/>Representing an average of true z coordinate values for all points;

When R ² is less than 0.95, returning to the step 3.4.1, and randomly selecting three points again to perform curve fitting until R ² is more than or equal to 0.95, and finishing the curve fitting;

Fitting P ₁ by using the steps 3.4.1 and 3.4.2 to obtain a curve segment function f ₁ (x), fitting P ₃ by using the steps 3.3.1 and 3.3.2 to obtain a curve segment function f ₂ (x);

And 3.5, repeating the steps 3.1 to 3.4, obtaining two straight line segment functions, two curve segment functions, the number of points in the point cloud slice data and region division in all passenger car types and the side wall plate point cloud slice data, and storing the fitted two straight line segment functions, two curve segment functions, the number of points in the point cloud slice data and region division and corresponding passenger car types into a template library.

5. The method for detecting the flatness of the side wall panel of the passenger car based on the three-dimensional point cloud and the contour matching according to claim 1, wherein the step 4 is specifically:

Step 4.1, acquiring three-dimensional point cloud data of a detected piece by utilizing a scanning device of a side wall board of the passenger car, and when the first frame of point cloud data is acquired, carrying out a preprocessing mode in the step 2.1 by a vertical horse, and simultaneously adopting a point cloud pose adjustment mode in the step 2.2 to enable Y coordinates of all points in the same frame of point cloud data to be the same;

step 4.2, calculating the number of points in the acquired first frame point cloud slice data, and finding the passenger car type closest to the calculated number of points in the first frame point cloud slice data in a template library and matching the passenger car type with the passenger car type;

Step 4.3, extracting two straight line segment functions and two curve segment functions corresponding to the matched passenger car type from a template library, dividing acquired point cloud slice data of the detected piece according to regions in the template library, dividing a set P corresponding to the point cloud slice data of the detected piece into four parts P ₁、P₂、P₃、P₄ according to x coordinates, wherein P ₁ and P ₃ are curve part point sets, P ₂ and P ₄ are straight line part point sets, and calculating the shortest distance from each point in each frame of the acquired point cloud slice data of the detected piece to the corresponding line segment function in the matched passenger car type;

Step 4.4, when obtaining the shortest distance from the point in the point set of the linear part of the measured object to the linear segment function, assuming that the linear segment function l ₁ (x) is converted into a linear equation ax+bz+c=0, the shortest distance from the point to the linear segment is:

Distance _i in equation (5) is the shortest distance from point (x _i,z_i) to straight line segment l ₁ (x);

Obtaining the shortest distance from each point in the P ₂ point set of the tested piece to a straight line segment function l ₁ (x) corresponding to the matched passenger car type in the template library by using the method of the step 4.4, and obtaining the shortest distance from each point in the P ₄ point set of the tested piece to a straight line segment function l ₂ (x) corresponding to the matched passenger car type in the template library;

Step 4.5, when obtaining the shortest distance from the point concentrated on the curve part of the measured object to the curve segment function, it is necessary to find the perpendicular point q ₀ between the point and the curve, assuming a point P _i(x_i,z_i concentrated on the point P ₁ of the measured object), taking a point q _τ on the curve segment function f ₁ (x), obtaining the direction vector of the tangent line at the point q _τ P _i(x_i,z_i) and q _τ are each a segment direction vector/>Setting a correlation threshold k, if/>Q _τ is the calculated point P _i(x_i,z_i) to the perpendicular point q ₀ between the curve segment functions f ₁ (x), otherwise, adding a step size, searching the next point on the curve segment function f ₁ (x), repeating the step of finding the perpendicular point until the point is satisfiedConditions of (2); after the perpendicular point q ₀ = (x, z) is found, the shortest distance from the point P _i(x_i,z_i to the curve segment function f ₁ (x) can be found:

Distance _i in equation (6) is the shortest distance from point P _i(x_i,z_i) to curve segment function f ₁ (x);

Calculating the shortest distance from each point in the P ₁ point set of the tested piece to a curve segment function f ₁ (x) corresponding to the matched passenger car type in the template library by using the method of the step 4.5, and calculating the shortest distance from each point in the P ₃ point set of the tested piece to a curve segment function f ₂ (x) corresponding to the matched passenger car type in the template library;

Step 4.6, setting a relevant defect threshold epsilon, respectively comparing the shortest distance between each point in the P ₂ point set of the tested piece and a straight line segment function l ₁ (x) corresponding to the matched passenger car type in the template library, the shortest distance between each point in the P ₄ point set of the tested piece and a straight line segment function l ₂ (x) corresponding to the matched passenger car type in the template library, the shortest distance between each point in the P ₁ point set of the tested piece and a curve segment function f ₁ (x) corresponding to the matched passenger car type in the template library, and the shortest distance between each point in the P ₃ point set of the tested piece and a curve segment function f ₂ (x) corresponding to the matched passenger car type in the template library with the relevant defect threshold epsilon, and if a certain shortest distance is larger than the relevant defect threshold epsilon, the point in the tested piece corresponding to the shortest distance is a defect point, and recording the shortest distance corresponding to the defect point and the position of the defect point;

and step 4.7, after the step 4.1 and the step 4.2 are performed, each frame of point cloud slice data of the measured piece is acquired, namely, the step 4.2 to the step 4.6 are circulated once.

6. The passenger car side wall panel flatness detection method based on three-dimensional point cloud and contour matching according to claim 1, wherein the step 5 specifically comprises:

and 5.1, when all the point cloud slice data of the side wall plate of the measured piece are acquired, expanding all the acquired point cloud slice data according to the Y direction to obtain the point cloud data of the whole side wall plate, and marking the defect area consisting of the defect points.

7. The passenger car side wall panel flatness detection method based on three-dimensional point cloud and contour matching according to claim 1, wherein the step 6 specifically comprises:

step 6.1, extracting all defect areas, and dividing the extracted defect areas into a defect area 1, a defect area 2 and a … defect area n;

step 6.2, assuming that the set of points in the defect area j is Q, j=1, 2,3 … n, calculating the minimum bounding box of all points in the set Q, obtaining the volume T of the minimum bounding sphere, setting a relevant threshold value gamma, and when T is less than gamma, recognizing the defect area as a misjudgment area and correcting the misjudgment area as a non-defect area;

And 6.3, detecting all the defect areas according to the step 6.2, removing the misjudgment areas in the defect areas, and finally obtaining the defect areas.