Blind zone point cloud self-adaptive complementing method and automatic welding seam tracing method applying blind zone point cloud self-adaptive complementing method

Abstract

The invention provides a blind area point cloud self-adaptive completion method and a welding seam automatic tracing method applying the same, and the method comprises the following steps: starting a camera, and collecting 3D point cloud data of a target object; processing the 3D point cloud data, and analyzing an area where a target blind area to be supplemented is located; the execution mechanism is controlled to drive the laser contourgraph to move to the position near the target blind area, the execution mechanism is controlled to drive the laser contourgraph to scan the target blind area, and contour data of the target blind area are obtained; and identifying the welding seam in the blind area according to the collected blind area data. According to the method provided by the invention, the area of the blind area caused by the limited visual field of the camera is analyzed, the blind area range is scanned by using the laser contourgraph, the blind area point cloud of the method is adaptively completed, and the complete feature information including the blind area is obtained by combining the global shooting of the camera with the local fine scanning of the contourgraph. And features in blind areas such as welding seams are automatically tracked based on complemented data.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of feature recognition, in particular to a blind area point cloud adaptive completion method and a welding seam automatic tracing method using the same. BACKGROUND

[0002] When a camera is used to identify the surface features of a target object, if there is a protrusion on the surface of the target object, the camera's line of sight will be limited, resulting in a blind area in the field of view. The features in the blind area lack camera data and cannot be identified.

[0003] The prior art installs multiple cameras at different positions and angles to ensure that the fields of view of these cameras can cover all sides of the protrusion, including the side that originally has a blind area. By fusing and stitching the image data collected by multiple cameras, complete facade information of the protrusion can be obtained, thereby solving the blind area problem. For example, patent document CN216747474U mentions using multiple fixed cameras to capture from multiple angles to obtain complete target features, or rotating the target object to obtain complete features. However, this approach has the problems of high cost, expensive price, complex design, and inflexible use. If the scene changes, the multi-camera layout needs to be changed accordingly.

[0004] Generally, cameras have a field of view and are relatively fixed focus, requiring a slight distance to capture the target object. For local blind area data, if a structured light camera is used for shooting, due to the long shooting distance and external light interference, the local fine structure may be lost. SUMMARY

[0005] The present application aims to solve the above problems and provides a blind area point cloud adaptive completion method and a welding seam automatic tracing method using the same. The complete feature information including the blind area is obtained through global shooting by a camera and local fine scanning by a profilometer, and the features in the blind area are automatically traced based on the completed information.

[0006] The present application solves the problem by adopting the following technical solution: a blind area point cloud adaptive completion method, comprising an execution mechanism, a laser profilometer, and a camera. The laser profilometer is installed at the end of the execution mechanism and moves under the driving of the execution mechanism. The method comprises the following steps: S1 Start the camera to collect 3D point cloud data of a target object; the target object is composed of a base plane and a protrusion on the base plane; S2 Process the 3D point cloud data to analyze the region of the target blind area to be filled; S3 Control the execution mechanism to drive the laser profilometer to move to the vicinity of the target blind area, and control the execution mechanism to drive the laser profilometer to scan the target blind area to obtain the profile data of the target blind area.

[0007] In step S2, the region where the target blind zone is located is analyzed using the following method: S2.1 Fitting the reference base plane F; S2.2 Project the points in the 3D point cloud data onto the reference base plane F; S2.3 Identify and filter out the point cloud empty areas in the projection points that need to be filled in the target blind area.

[0008] The area where the target blind zone is located is calculated using the following method: Extract the set A{A1,A2,......A1} of the protrusion edge points from the feature point cloud on the upper surface of the protrusion, near the target blind zone. m}; i = 1, 2, ..., m; Get the edge point A of the protrusion i The intersection point B of the line connecting the camera coordinate system origin O and the reference base plane F. i Obtain the set of the base edges of the blind zone B{B1,B2......B m}; The set of protrusion edge points A and the set of blind zone base edges B are projected onto the reference base plane F, and the contour formed by the projection points is the area where the target blind zone is located.

[0009] The pose of the laser profilometer scanning the target blind zone is obtained using the following method: Let the set of edge points of the protrusion be A{A1,A2,......A...} m The points in} are projected onto the reference base plane F to obtain the point cloud set Ap{Ap1,Ap2,......Ap}. m}; On the reference base plane F, calculate the contour line of the point cloud set Ap at point Ap. i The intersection point Ap' of the normal line at the location and the edge line of the blind zone base. i The blind zone base edge line is obtained by fitting points in the blind zone base edge set B; Calculate point A i With point Ap' i vector Normalized vector For the posture of the laser profilometer ; Calculate point A i Ap' i The unit vector perpendicular to the straight line is used as the attitude of the laser profilometer. ; Calculate point A i With point Ap' i Distance DIi The minimum fill width region of the laser profilometer is defined, and the minimum height H of the laser profilometer origin corresponding to the minimum fill width region is calculated. imin The height H of the laser profilometer is obtained according to the preset tolerance. i ; Calculate point A i Ap' i Midpoint C of the straight line i The midpoint C i along Move the height H along the axis and away from the area where the target blind zone is located. i Obtain the origin position P of the laser profilometer. i .

[0010] The contour data of the target blind area obtained in step S3 is stitched together with the 3D point cloud data obtained by the camera, and point cloud empty areas are searched in the stitched data; if point cloud empty areas exist in the area where the target blind area is located and the surrounding area, the actuator is controlled to drive the laser profilometer to adjust its posture and rescan the target blind area.

[0011] Before data stitching, the overlap between the non-blind zone data collected by the laser profilometer and the 3D point cloud acquired by the camera is adjusted.

[0012] An automatic weld seam tracking method, applying the aforementioned blind zone point cloud adaptive completion method, includes the following steps: The contour data of the target blind zone collected by the laser profilometer is processed to obtain the target features within the blind zone; the target features within the blind zone include: a set of weld feature points L{L1,L2,......L...} formed by the protruding surface and the base plane within the target blind zone. m}; The laser profilometer is placed in the same scanning posture T i The obtained protrusion facade outline AL i and the base plane outline BL i The intersection point is used as the weld feature point L i .

[0013] The protrusion's facade outline AL i According to the posture T i The protrusion facade contour point set is obtained by fitting the data from the lower scan. The base plane contour line BL i According to the posture T i The base plane contour point set is obtained by fitting the data during the downward scan.

[0014] For the attitude T iThe acquired contour data point set, calculate the point Q in the contour data point set ij The distance dist to the reference base plane F ij ; j = 1, 2,..., R i ; filter dist ij Points < first threshold value, as the base plane contour point set; filter dist ij > second threshold value, as the convex facade contour point set.

[0015] The beneficial effects of the present application: by acquiring the point cloud data of the target object by the camera, the region of the blind area caused by the limited field of view of the camera is actively analyzed, and the laser profiler is used to scan the blind area range, the blind area point cloud is adaptively completed, the complete feature information including the blind area is obtained by combining the global shooting of the camera with the local fine scanning of the profiler, and the problem that the surface feature data cannot be completely collected due to the characteristics such as the convex object contained in the target object is solved; further, according to the adaptively completed point cloud information, the characteristics such as the weld in the blind area are automatically tracked.

[0016] The laser profiler is small in size and flexible in attitude adjustment, and can be flexibly applied to different scenes to complete the blind area data; by using the method, the laser profiler can be directly carried on the working robot for cooperation, without special structure design, installation and simple operation and low cost.

[0017] The laser profiler scans the blind area at a short distance, which is less susceptible to external light interference than traditional structured light shooting, and is more suitable for collecting obvious convex and concave profiles, effectively reducing data loss or distortion.

[0018] In order to obtain the local fine structure of the blind area, the pose of the laser profiler needs to be close to the blind area, and at the same time needs to maintain a certain distance from the blind area to avoid incomplete shooting, the present application proposes a laser profiler scanning pose construction method for the blind area in the case of lacking blind area data, and automatically constructs the best scanning pose of the laser profiler, which is close enough to the blind area while covering the blank area to be completed, and ensures the quality of the acquired blind area feature data.

[0019] By using the conversion relationship between different coordinate systems, the laser profiler scanning data and the camera acquisition data are spliced to obtain the feature information of the target object including the blind area, and the spliced data is verified, the laser profiler pose is adjusted according to the data loss condition to be scanned again, and the integrity of the finally acquired and spliced data is ensured.

[0020] The present application proposes an automatic identification and screening method for the target blind area, which has high identification accuracy and can effectively avoid the interference of environmental light and other factors on the blind area identification.

[0021] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the application. BRIEF DESCRIPTION OF DRAWINGS

[0022] The accompanying drawings incorporated in and forming a part of the specification illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the application.

[0023] Figure 1 Flow chart of the adaptive blind area point cloud completion method in the application; Figure 2 Schematic diagram of the implementation environment of the application; Figure 3 Schematic diagram of the adaptive blind area identification, acquisition and splicing system in an embodiment of the application; Figure 4 Schematic diagram of obtaining the pose of the laser profiler in an embodiment of the application. DETAILED DESCRIPTION

[0024] To make the above objectives, features and advantages of the present application more apparent, more comprehensible and more understandable, the present application will be further described in detail below with reference to the drawings and specific embodiments.

[0025] An adaptive blind area point cloud completion method, the implementation environment is as shown in Figure 2 The implementation environment includes an adaptive blind area identification, acquisition and splicing system 1 and a target object 2, and the adaptive blind area identification, acquisition and splicing system is as shown in Figure 3 It includes an execution mechanism 11, a laser profiler 12, a camera 13 and a control system. The laser profiler 12 is installed at the end of the execution mechanism. Under the control of the control system, the execution mechanism 11 drives the laser profiler 12 to move. The execution mechanism can adopt a mechanical arm, a mechanical hand, etc.

[0026] In a specific embodiment, the implementation scenario of the present method is a large-size scene. The camera is installed on a fixed platform position. The approximate information of the target object is obtained through double cameras.

[0027] The target object is composed of a base plane and protrusions on the base plane. Due to the influence of the protrusions, there is a certain blind area in the field of view of the camera. In the field of view of the camera, the points on the base plane around the back vertical surface of the protrusions cannot be collected, forming a blind area.

[0028] The adaptive blind area point cloud completion method is as shown in Figure 1 It includes the following steps: S1 Starting the camera, collecting 3D point cloud data of the target object; S2 Processing the 3D point cloud data, analyzing the region of the target blind area to be filled; S3 controls the execution mechanism to drive the laser profiler to move to the vicinity of the target blind area, controls the execution mechanism to drive the laser profiler to scan the target blind area, and obtains the profile data of the target blind area.

[0029] The target blind area is analyzed in the following method in the S2 step: S2.1 fitting a reference base plane F; S2.2 projecting the points in the 3D point cloud data onto the reference base plane F; S2.3 identifying and screening the point cloud empty area of the target blind area to be filled in the projected points.

[0030] The reference base plane F can be fitted by using the method of statistics according to the depth information: the base plane has a large number of point clouds and has a similar depth value. The depth information is used to statistically analyze all point clouds, and then the OTSU method is used for binary classification, and the class with more point clouds is the point cloud of the base plane, and the base plane is fitted by using the point cloud.

[0031] The method of demarcating the area of interest can be used: specifically, the position where the base plane may appear in the 2D image is demarcated, and the demarcated range is appropriately reduced to ensure that the base plane does not contain other elements. The 2D image in the demarcated position is mapped to the 3D point cloud, and the base plane is fitted by using the point cloud.

[0032] In the S2.3 step, the point cloud empty area of the target blind area to be filled can be screened by the following method: In one specific embodiment, the approximate range of the target blind area to be filled is manually framed, and the point cloud empty area is identified as the target blind area to be filled in the approximate range manually framed.

[0033] In another embodiment, the target blind area to be filled is analyzed by the positional relationship between the target blind area and the point cloud empty area; the target blind area is first calculated and enlarged, and the point cloud empty area is identified as the target blind area to be filled in the enlarged target blind area.

[0034] In another embodiment, the point cloud empty area is first identified in the projected points, the coincidence degree between the point cloud empty area and the target blind area is analyzed, and the target blind area to be filled is selected by the coincidence degree.

[0035] In one specific embodiment, the target blind area is calculated by the following method: The edge point set A {A1, A2,..., Am} of the convex edge of the convex object on the surface feature point cloud of the target object near the target blind area is extracted. m};i=1,2......m; In one specific embodiment, the edge point set A is obtained by the following method: Define the region of interest (ROI), traverse each 3D point in the ROI, and search for 3D points in its 8-neighborhood. If 3 of the 3D points in the 8-neighborhood are empty, then the point is considered to be the edge point of the protrusion near the blind zone of the target.

[0036] In another embodiment, the method for obtaining the edge point set A is as follows: First, for the 3D point cloud data acquired by the camera in step S1, filter out data whose distance from the reference base plane F is greater than a threshold T. x The point, threshold T x It can be set to 1 / 2 the height of the upper surface of the protrusion from the base plane, resulting in a point set U. For each point in the point set U, search for 3D points in its 8-neighborhood. If 3 of the 3D points in the 8-neighborhood are empty, then the point is considered to be the edge point of the protrusion near the blind zone of the target.

[0037] Get the edge point A of the protrusion i The intersection point B of the line connecting the camera coordinate system origin O and the reference base plane F. i Obtain the set of the base edges of the blind zone B{B1,B2......B m}; Project the set of protrusion edge points A and the set of blind zone base edges B onto the reference base plane F. The contour formed by the projected points is the area where the target blind zone is located.

[0038] In one specific embodiment, all points in the protrusion edge point set A and all points in the blind zone base edge set B are projected onto the reference base plane F, and the contour formed by the projected points is the area where the target blind zone is located.

[0039] After obtaining the location of the target blind zone, the actuator drives the laser profilometer to scan the target blind zone and obtain the contour data of the target blind zone. In order to ensure that the collected contour data is clear and effective, the laser profilometer needs to acquire the data in an appropriate pose.

[0040] In one specific embodiment, the pose of the target blind area scanned by the laser profilometer is obtained using the following method: like Figure 4 As shown, the set of edge points A{A1,A2,......A... of the protrusion is... m The points in} are projected onto the reference base plane F to obtain the point cloud set Ap{Ap1,Ap2,......Ap}. m}; On the reference base plane F, calculate the contour line of the point cloud set Ap at point Ap. i The intersection point Ap' of the normal line at the location and the edge line of the blind zone base. i The blind zone base edge line is obtained by fitting points in the blind zone base edge set B; The following method establishes the projection posture of the laser profiler: The vector of point A i and point Ap' i ; The normalized vector is the of the laser profiler posture; The perpendicular direction unit vector of the straight line from point A i to point Ap' i is the of the laser profiler posture; , Cross product to obtain ; The following method establishes the origin position of the laser profiler: The distance DI i between point A i and point Ap' i is the minimum supplementary width area of the laser profiler, and the corresponding minimum height H imin of the laser profiler origin covering the minimum supplementary width area is calculated, and the profile line projected from the collection height needs to cover at least the minimum supplementary width DI i , and the height H i of the profiler is obtained according to the preset tolerance; The midpoint C i of the straight line from point A i to point Ap' i , move the midpoint C i along the axis and in the direction away from the area where the target blind area is located by H i , to obtain the origin position P i of the laser profiler; as Figure 4 shown , when P i points to the target blind area, C i = P i -H i × ; if P i is away from the target blind area, C i = P i +H i × ; For each point A i , take P imin as the origin, and , establish the corresponding laser profiler pose W i , obtain the pose set W of the laser profiler scanning blind area, and drive the laser profiler along the pose set W to collect blind area profile data.

[0041] For the profile data of the target blind area obtained by the above steps, the profile data is further spliced with the 3D point cloud data obtained by the camera to obtain global camera data and blind area fine data.

[0042] Before data splicing, for the non-blind area data collected by the laser profiler, the coincidence degree of the non-blind area data and the 3D point cloud obtained by the camera is adjusted.

[0043] Since the laser profiler scan sets a preset tolerance when establishing the pose, a part of non-blind area data can be collected, so that through the coincidence of the non-blind area data, a part of adaptive pose adjustment can be made when splicing to the camera coordinate system, so as to reduce the splicing error caused by inaccurate hand-eye calibration of the profiler.

[0044] The camera acquisition data and the laser profiler scan data both contain protrusion edge features, so the protrusion edge features can be used as reference objects to compare the coincidence degree between the camera data and the laser profiler data and to splice them. In the process, ICP algorithm can be used for optimization correction.

[0045] For the spliced data, the point cloud empty area is searched in the spliced data; If the target blind area and the nearby area still have point cloud empty areas, the control executes the mechanism to drive the laser profiler to adjust the pose and re-scan the target blind area.

[0046] One implementation is to obtain the collection height H imin of the profiler based on the minimum height H i and the preset tolerance, recalculate the new height H i by increasing the preset tolerance, and make the profile line of the profiler cover more width.

[0047] Another implementation is to regard the incomplete point cloud empty area as a new blind area, remove the points in the completed area in the above protrusion edge point set A from the set, take the remaining points as a new protrusion edge point set A', remove the points in the completed area in the above blind area base edge set B from the set, take the remaining points as a new blind area base edge set B', and repeat the above steps for the new set A' and set B'. This implementation is suitable for the case where the incomplete point cloud empty area is scattered. If necessary, the collection height of the profiler can also be increased for the new set to make the profile line of the profiler cover more width.

[0048] The above method addresses the situation where the unfilled area is located at the edge of the original blind zone. Another scenario is where non-edge areas within the original blind zone are not filled, creating empty point cloud areas within the original blind zone. This could be due to factors such as reflection. A feasible solution for this situation is: on the fitted reference base plane F, subtract the points within the filled area from the original blind zone to obtain a new blind zone. For this new blind zone, extract its edge points, and construct a local geometric model based on geometric features such as the point cloud's normal or curvature to predict and fill in the empty areas.

[0049] The data stitching method between camera acquisition data and laser profilometer scanning data is as follows: If each frame of line laser data can correspond one-to-one with the robot arm coordinates and the accuracy meets the requirements, then each frame of line laser data (each frame is a line) can be transferred to the robot arm base coordinate system according to the robot arm coordinates and calibration results, and then transferred to the camera coordinate system or standard coordinate system.

[0050] Another implementation method is to transfer each frame of line laser data to the laser profilometer coordinate system at the scanning starting point based on the starting position of the laser profilometer, the scanning direction, and the moving speed of the actuator. Then, based on the starting point robot arm coordinates, the data is transferred to the robot arm base coordinate system, and then to the camera coordinate system or standard coordinate system.

[0051] In one implementation scenario, target features are further extracted from the contour data collected by the laser profilometer.

[0052] Based on this, an automatic weld seam tracking method is proposed. This method processes the contour data of the target blind zone to obtain the target features within the blind zone. The target features of the blind zone include: a set of weld seam feature points L{L1, L2, ..., L...} formed by the protruding surface and the base plane within the blind zone. m}; The laser profilometer is placed in the same scanning posture T i The obtained protrusion facade outline AL i and the base plane outline BL i The intersection point is used as the weld feature point L i .

[0053] AL of the protruding facade i According to posture T i The protrusion facade contour point set is obtained by fitting the data from the lower scan. Base plane contour line BL i According to posture T i The base plane contour point set is obtained by fitting the data during the downward scan.

[0054] For posture T i The obtained contour data point set is used to calculate point Q in the contour data point set. ij Dist to the reference base plane F ijj = 1, 2,..., R i ; Screening dist ij Points < first threshold value, as the base plane contour point set; Screening dist ij Points > second threshold value, as the convex facade contour point set.

[0055] The second threshold value can be set to a value greater than the first threshold value, set according to the assembly tolerance and identification error.

[0056] Compared with the prior art, the application actively analyzes the area of the blind area caused by the limited field of view of the camera by collecting point cloud data of the target object by the camera, and scans the blind area range by the laser profiler to adaptively complete the blind area point cloud. The complete feature information including the blind area is obtained by combining the global shooting of the camera with the local fine scanning of the profiler, solving the problem that the surface feature data cannot be completely collected due to the characteristics such as convex objects contained in the target object. Further, according to the adaptively completed point cloud information, the features such as welds in the blind area are automatically tracked. The laser profiler is small in size and flexible in attitude adjustment, and can be flexibly applied to different scenes to fill in the blind area data. By using the method, the laser profiler can be directly carried on the working robot for cooperation, without special structure design, installation and simple operation and low cost. By using the laser profiler to scan the blind area at a close distance, it is less susceptible to external light interference compared with traditional structured light shooting, and is more suitable for collecting profiles with obvious concave and convex features, effectively reducing data loss or distortion. In order to obtain the local fine structure of the blind area, the pose of the laser profiler needs to be close to the blind area and at the same time needs to maintain a certain distance from the blind area to avoid incomplete shooting. The application proposes a laser profiler scanning pose construction method for the blind area in the case of lacking blind area data, automatically constructing the best scanning pose of the laser profiler, which is close enough to the blind area while ensuring the coverage of the blank area to be filled in, and ensures the quality of the acquired blind area feature data. By using the conversion relationship between different coordinate systems, the laser profiler scanning data and the camera acquisition data are spliced to obtain the feature information of the target object including the blind area, and the spliced data is verified. According to the data loss condition, the laser profiler pose is adjusted and scanned again to ensure the integrity of the finally acquired and spliced data. The application also proposes an automatic identification and screening method for the target blind area, which has high identification accuracy and can effectively avoid the interference of environmental light and other factors on the blind area identification.

[0057] In addition, it should be understood that the above embodiments only express several embodiments of the application, and the description is more specific and detailed, but it should not be understood as limiting the scope of the patent. It should be noted that for ordinary skilled in the art, without departing from the concept of the application, a number of modifications and improvements can be made, which are within the scope of protection of the application.

Claims

1. A method for blind area point cloud adaptive completion, comprising an execution mechanism, a laser profiler and a camera, the laser profiler is installed on the end of the execution mechanism and moves under the driving of the execution mechanism; characterized in that, Comprising the following steps: S1 Start the camera to collect 3D point cloud data of the target object; the target object is composed of a base plane and a convex object on the base plane; S2 Process the 3D point cloud data to analyze the area where the target blind area to be filled is located; S3 Control the execution mechanism to drive the laser profiler to move to the vicinity of the target blind area, and control the execution mechanism to drive the laser profiler to scan the target blind area to obtain the contour data of the target blind area.

2. The blind area point cloud adaptive completion method of claim 1, wherein, The following method is used in the S2 step to analyze the area where the target blind area is located: S2.1 Fit a reference base plane F; S2.2 Project the points in the 3D point cloud data onto the reference base plane F; S2.3 Identify and select the point cloud empty area of the target blind area to be filled among the projected points.

3. The blind area point cloud adaptive completion method of claim 1, wherein, The area where the target blind area is located is obtained by the following method: extracting a set of convex edge points A {A1, A2,... A} close to one side of the target blind area in the upper surface feature point cloud of the convex object m}; i = 1, 2,... m; Obtaining the convex edge point A i The intersection point B of the line connecting the camera coordinate system origin O and the reference base plane F i , obtaining the blind area base edge set B{B1, B2......B m} Project the convex object edge point set A and the blind area base edge set B onto the reference base plane F, and the contour formed by the projected points is the area where the target blind area is located.

4. The blind area point cloud adaptive completion method of claim 3, wherein, The following method is used to obtain the pose of the laser profiler scanning the target blind area: Let the set of edge points of the protrusion be A{A1,A2,......A...} m The points in} are projected onto the reference base plane F to obtain the point cloud set Ap{Ap1,Ap2,......Ap}. m }; On said reference base plane F, the intersection Ap' of the normal straight line to the point cloud set Ap profile line at point Ap i , obtained according to the points in the blind zone base edge set B, is calculated i . Computing point A i Vector to point Ap' i Vector to point Ap' Normalized vector For laser profiler pose For laser profiler pose Computing point A i To point Ap' i The perpendicular direction unit vector of the straight line as the laser profiler pose ; Computing point A i Distance DI of point Ap' from point A i i The minimum supplementary width region of the laser profiler, and computing the minimum height H of the laser profiler origin point corresponding to the minimum supplementary width region imin According to the preset tolerance, the height H of the laser profiler is obtained i ;​ Point A i to point Ap' i Midpoint C of the straight line i , the midpoint C i along axis and away from the target blind area region direction of the height H i , get the origin position of the laser profiler P i .

5. The blind area point cloud adaptive completion method of claim 4, wherein, Splice the contour data of the target blind area obtained in the step S3 with the 3D point cloud data obtained by the camera, and find the point cloud empty area in the spliced data; if there is a point cloud empty area in the area where the target blind area is located and the surrounding area, control the execution mechanism to drive the laser profiler to adjust the posture and rescan the target blind area.

6. The blind area point cloud adaptive completion method of claim 5, wherein, Before data splicing, adjust the coincidence degree of the non-blind area data collected by the laser profiler and the 3D point cloud data obtained by the camera.

7. A method of automatic seam tracking, characterized in that A blind area point cloud adaptive completion method as claimed in claims 1-6, comprising the following steps: Processing the profile data of the target blind area collected by the laser profiler to obtain a target feature in the blind area; the target feature in the blind area includes a set of weld feature points L{L1, L2,..., L m} formed by the standing surface of the convex and the base plane in the target blind area; and taking the intersection of the standing surface profile line AL i and the base plane profile line BL i obtained by the laser profiler in the same scanning posture T i as the weld feature point L i .

8. A method of automatic seam tracking as claimed in claim 7, characterized in that The convexity facade profile line AL i According to the posture T i The convexity facade profile point set of the lower scan is obtained by fitting; and the base plane profile line BL i According to the posture T i The base plane profile point set of the lower scan is obtained by fitting.

9. A method of automatic seam tracking as claimed in claim 8, characterized in that For the pose T i The set of profile data points acquired next, calculate the points Q in the set of profile data points ij The distance dist to the reference base plane F ij ; j = 1, 2... R i ; filter the points with dist ij < the first threshold value, as the set of base plane profile points; filter the points with dist ij > the second threshold value, as the set of convex facade profile points.

Images