3d vision and robot collaborative precision detection method for multiple types of steel bars

Abstract

This invention discloses a 3D vision and robot collaborative precision detection method for multiple types of reinforcing bars. This method utilizes a 3D area array camera to acquire point cloud and texture data of the reinforcing bar area, sequentially performing background plane detection, reinforcing bar point cloud segmentation, multi-view stitching, model registration, and size detection. Robust planar benchmark construction is achieved through adaptive image enhancement and depth constraint filtering algorithms. A circular marker-based SVD registration algorithm is used to achieve full-view point cloud stitching for ultra-long reinforcing bars. Combined with a 360° rotation search and type-based detection strategy, automatic 3D pose and size measurement of straight, bent, and curved reinforcing bars is achieved. This invention solves the problems of traditional 2D vision's inability to acquire depth information and the low efficiency of manual measurement, realizing fully automated detection of multiple types of reinforcing bars. The detection cycle is only 5-10 seconds, the stitching accuracy is better than 2mm, and pose parameters can be directly output for robot collaborative operation.

Description

Technical Field

[0001] This invention belongs to the field of industrial automation inspection and robotics technology, specifically relating to a 3D vision and robot collaborative precision inspection method for multiple types of steel bars. Background Technology

[0002] In building construction and underground shield tunnel engineering, rebar tying is a crucial step in reinforced concrete structure construction. Its quality directly affects the load-bearing capacity, overall safety, and long-term durability of the main structure, and has a significant impact on shield tunnel segment assembly, structural stress, and subsequent operational safety. Traditional rebar inspection mainly relies on manual visual inspection, which has the following significant problems:

[0003] Inefficient: Manual inspection is slow and cannot meet the needs of large-scale construction progress.

[0004] Highly subjective: The test results are greatly affected by the experience and fatigue level of the testers, resulting in a high rate of missed detections and false detections.

[0005] Poor environmental adaptability: The lighting at the construction site is complex and the steel bars are piled up in a mess, making it difficult for people to fully observe the hidden parts.

[0006] Difficulty in adapting to multiple types: There is a lack of unified automated detection standards for steel bars with different diameters, spacings, and binding methods. Existing single vision solutions are difficult to accurately identify multiple types of steel bars.

[0007] While some 2D vision-based detection attempts have been made, limitations in lighting variations and lack of depth information prevent accurate determination of the spatial location, overlap, and 3D coordinates of the ties of the reinforcing bars. This hinders precise subsequent operations or quantitative analysis by the robot. Therefore, a collaborative robot detection method with high-precision 3D perception capabilities, adaptable to various types of reinforcing bar scenarios, is urgently needed. Summary of the Invention

[0008] The purpose of this invention is to provide a 3D vision and robot collaborative precision detection method for multiple types of steel bars, so as to solve the problems mentioned in the background art.

[0009] To achieve the above objectives, the present invention provides the following technical solution: a method for precise detection of multiple types of reinforcing bars using 3D vision and robotic collaboration, comprising the following steps:

[0010] S1: 3D point cloud acquisition, using a 3D area array camera to acquire the original point cloud data and corresponding 2D texture image of the area of ​​the steel bar to be inspected;

[0011] S2: Background plane detection: Based on the 2D texture image and the original point cloud data, determine the workbench plane where the reinforcing bars are placed and construct the background plane reference.

[0012] S3: Rebar point cloud segmentation: Based on the background plane reference, extract independent point cloud instances of each rebar from the original point cloud data;

[0013] S4: Multi-view point cloud stitching. If the length of the steel bar exceeds the field of view of a single camera, the point cloud is registered and stitched by acquiring data from multiple perspectives and using preset markers to generate a complete steel bar point cloud.

[0014] S5: Rebar model registration, which aligns the extracted or spliced ​​rebar point cloud with the pre-stored standard model to determine the three-dimensional pose of the rebar;

[0015] S6: Rebar size inspection. In the registered local coordinate system, select the corresponding inspection strategy according to the shape type of the rebar and calculate the size parameters of the rebar.

[0016] S7: Output the results, outputting the three-dimensional pose and dimensional parameters of the reinforcing bar to the robot control system.

[0017] Preferably, step S2: background plane detection specifically includes:

[0018] (1) Image enhancement: Select adaptive limited contrast histogram equalization (CLAHE) or multi-scale Retinex (MSR) algorithm according to the size of the platform to enhance the grayscale contrast between the platform and the steel reinforcement area;

[0019] (2) Adaptive threshold segmentation: The OTSU algorithm is used to automatically determine the threshold for the enhanced grayscale image and generate a planar candidate mask;

[0020] (3) Morphological refinement: Perform closing and dilation operations on the planar candidate mask in sequence to obtain the refined planar region mask;

[0021] (4) Depth constraint filtering: Combine the Z-axis depth value of the point cloud, retain only the points whose depth is within the preset window of the target platform height, and remove the points with abnormal depth;

[0022] (5) Least squares plane fitting: The SVD decomposition is used to perform least squares plane fitting on the selected candidate point set of the plane, solve the normal vector and origin of the plane equation, and output the background plane.

[0023] Preferably, step S3: rebar point cloud segmentation specifically includes:

[0024] (1) Planar distance extraction: Calculate the distance from each valid 3D point to the background plane, and retain the points whose distance is within the preset range to generate a set of candidate reinforcement points;

[0025] (2) Projection binary image generation: Map the candidate point set of steel bars back to 2D image coordinates to generate a candidate region map, and perform dilation operation to connect the fracture region;

[0026] (3) Connected component analysis and instance segmentation: Perform connected component analysis, assign an independent label to each connected component, filter out noisy areas and incomplete edge areas, and each connected component that passes the filter corresponds to an independent steel bar;

[0027] (4) 3D instance point cloud recovery: Based on the pixel coordinates of the 2D connected domain, extract the corresponding 3D coordinates from the original point cloud array to obtain an independent 3D point cloud set for each steel bar.

[0028] Preferably, step S4: multi-view point cloud stitching specifically includes:

[0029] (1) Circular marker detection: In the 2D texture images of multiple cameras, detect the circular calibration plate pre-set on the worktable and obtain the center coordinates of the marker pixels;

[0030] (2) Marker 3D coordinate extraction: For each detected circular marker, the effective 3D points near its pixel center are averaged to accurately extract the three-dimensional spatial coordinates of the marker;

[0031] (3) SVD point-to-point registration: Using the 3D point pairs of markers that are visible to each other in adjacent views as input, the optimal rigid body transformation matrix is ​​solved by SVD decomposition, and the point clouds of multiple views are merged into the same coordinate system to complete the stitching.

[0032] Preferably, step S5: rebar model registration specifically includes:

[0033] (1) Construction of local coordinate system: With the background plane normal vector as the Z-axis, the plane projection point of the centroid of the steel reinforcement point cloud as the origin, and the projection direction of the centroid pointing to the first skeleton point as the X-axis, a right-handed local coordinate system is constructed.

[0034] (2) Global rotation search registration: Transform the pre-stored standard model skeleton point set to the local coordinate system, traverse all rotation angles with a preset step size, including the case of rotation around the axis, calculate the average distance from the measured skeleton point to the nearest model point, and select the transformation matrix corresponding to the minimum average distance as the final registration result.

[0035] Preferably, in step S6: rebar size inspection, different inspection strategies are adopted for different types of rebar:

[0036] When the reinforcement is a straight line: fit a straight line in space, take the first and last two points of the effective point cloud, calculate the distance to obtain the length;

[0037] When the reinforcement is a rectangular frame: classify the skeleton point set into four model reference lines, extract the farthest endpoint pair to obtain the four vertices of the rectangle, calculate the distance between each vertex to obtain the length of each side, and determine the starting end by the point cloud density;

[0038] For L / U-shaped bent steel bars: perform linear fitting on the point cloud of each steel bar segment, calculate the spatial intersection of two adjacent straight lines to obtain the coordinates of the bend vertex, and calculate the length of each segment with the bend vertex as the boundary;

[0039] When the reinforcement is curved: take the coordinates of the centroid of the registered skeleton as the reference point of the geometric center, calculate the distance between each two adjacent points and sum them to obtain the length of the curve.

[0040] Preferably, the method is applied to robot collaborative operation scenarios:

[0041] The robot's end effector is equipped with the 3D area array camera, or the camera is carried by a mobile platform (AGV) and moves in coordination with the robotic arm;

[0042] In step S1, the robot or mobile platform moves to different camera positions according to a preset path to collect data;

[0043] After step S7, the robot control system performs subsequent grasping, sorting or processing operations based on the output three-dimensional pose and size parameters.

[0044] Preferably, in the image enhancement step: for long platforms, the CLAHE algorithm is used with parameters set as follows: clipLimit=3.0, GridSize=11×11; for small platforms, the multi-scale Retinex (MSR) algorithm is used, with Gaussian scale σ∈{15, 80, 250} decomposed in the logarithmic domain.

[0045] Preferably, in the planar distance extraction step, the preset range is 5~30mm from the background plane; in the connected component analysis step, noise areas with an area of ​​less than 3000 pixels are filtered out, and areas with a horizontal centroid located within 15% of the left and right edges of the image are excluded.

[0046] Preferably, the parameters of the circular calibration plate are set as follows: area range of 10,000~20,000 pixels, circularity of 0.75~1.0; the markers in the intermediate view are grouped left and right according to the horizontal center line of the image to assist in multi-camera stitching.

[0047] The technical effects and advantages of this invention are as follows:

[0048] 1. This invention uses a 3D area array camera combined with a background plane detection algorithm. Through least squares plane fitting and depth constraint filtering, it effectively eliminates lighting variations and metal reflection noise in industrial sites. It directly acquires the three-dimensional spatial point cloud data of steel bars, avoiding projection errors and improving the size detection accuracy by about an order of magnitude.

[0049] 2. This invention designs a categorized size detection strategy (covering straight, rectangular, L / U-shaped bends, and curved steel bars), combined with a model registration algorithm based on 360° rotation search, which can automatically identify and adapt to steel bars of different diameters, shapes, and arrangement states, achieving "one system for multiple uses" and solving the problem of poor adaptability to multiple types of steel bars in the prior art.

[0050] 3. For large steel reinforcement components, this invention proposes a multi-view point cloud stitching technology based on circular markers. By arranging a circular calibration plate on the workbench and using the SVD point-to-point registration algorithm, the local point clouds collected by multiple cameras are seamlessly stitched together, expanding the effective measurement field of view from about 1.2 meters of a single camera to more than 3.5 meters. The stitching accuracy is controlled within 2 mm, realizing the complete measurement of ultra-long steel reinforcement with full field of view.

[0051] 4. This invention achieves full automation of the entire process from point cloud acquisition, processing, registration to size calculation and result output, with a detection cycle of only 5 to 10 seconds. At the same time, the three-dimensional pose and size parameters output by this method can be directly connected to the robot control system to guide the robotic arm to grasp, sort or process, realizing unmanned and intelligent operation of the production line.

[0052] 5. By introducing multiple preprocessing mechanisms such as CLAHE / MSR image enhancement, OTSU adaptive threshold segmentation, and morphological refinement, this invention can still operate stably in industrial environments with complex backgrounds and uneven lighting, accurately extracting independent steel bar instances and ensuring the long-term operational stability of the system. Attached Figure Description

[0053] Figure 1 This is a flowchart of the present invention;

[0054] Figure 2 This is a schematic diagram of the splicing marker of the present invention. Detailed Implementation

[0055] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0056] Example 1: 3D visual inspection process for various types of reinforcing bars

[0057] This embodiment details the complete execution process of a 3D vision and robot collaborative precision detection method for various types of reinforcing bars. For example... Figure 1 As shown, this method is based on a collaborative platform of a 3D area scan camera and a robot, and the specific steps are as follows:

[0058] 1. 3D point cloud acquisition (S1)

[0059] A high-precision 3D area array camera (structured light camera) was used to photograph the area of ​​the reinforcing steel bars placed on the worktable. Raw point cloud data P containing depth information was acquired. raw and the corresponding 2D texture grayscale image I gray If the length of the steel bar to be measured exceeds the field of view of a single camera (approximately 1.2m), the multi-view stitching process (S4) will be initiated; otherwise, subsequent processing will proceed directly.

[0060] 2. Background plane detection (S2)

[0061] This step aims to determine the equations of the workbench plane, which will serve as the reference for subsequent segmentation. Specifically, it includes:

[0062] Image enhancement: Determine the size of the tabletop. If it is a long tabletop, use the CLAHE algorithm (parameters: clipLimit=3.0, GridSize=11×11) to enhance the contrast; if it is a small tabletop, use the multi-scale Retinex (MSR) algorithm (Gaussian scale σ∈{15,80,250}) to enhance it and eliminate uneven lighting.

[0063] Thresholding and Morphological Processing: The OTSU algorithm is used to automatically calculate the threshold, and the image is binarized to generate a planar candidate mask. Then, a 13×13 rectangular kernel closing operation and a 21×21 rectangular kernel dilation operation are performed sequentially to fill holes and expand region boundaries.

[0064] Depth filtering and plane fitting: Outliers with depth values ​​outside the target platform height ±70mm window are removed. For the remaining set of candidate plane points {P}... i The least-squares plane is fitted using SVD decomposition to solve for the normal vector n and origin P of the plane equation, and outputs the background plane Plane. bg .

[0065] 3. Reinforcement point cloud segmentation (S3)

[0066] Based on the background plane bg Extracting individual rebar examples:

[0067] Distance filtering: Calculate the distance (dist) from each valid 3D point to the background plane. Retain points with a distance between 5mm and 30mm to generate a candidate set of rebar points (removing background noise and excessively tall foreign objects).

[0068] Projection and Connectivity Analysis: The candidate point set is mapped back to 2D image coordinates to generate a binary candidate region map, and a 5×5 rectangular kernel dilation is performed to connect broken regions. An 8-connectivity analysis is performed to filter out noise regions with an area less than 3000 pixels and to exclude edge regions whose horizontal centroids are located within 15% of the left and right edges of the image.

[0069] Instance recovery: Each filtered connected component corresponds to an independent rebar. Based on the 2D connected component coordinates, the corresponding 3D coordinates are extracted from the original point cloud to obtain the independent point cloud set {P} for each rebar. bari}

[0070] 4. Multi-view point cloud stitching (S4, optional)

[0071] For large steel reinforcement components exceeding 3 meters in length:

[0072] Marker detection: A circular calibration plate is pre-placed on the worktable. In the 2D texture images from the left and right cameras, circular markers with an area of ​​10,000 to 20,000 pixels and a circularity of 0.75 to 1.0 are detected, and their pixel center coordinates are obtained.

[0073] 3D coordinate extraction: Take the average of the effective 3D points near the pixel center to accurately extract the three-dimensional spatial coordinates of the marker.

[0074] SVD registration: Using 3D point pairs of markers commonly visible in adjacent views as input, the optimal rigid body transformation matrix T is solved using SVD decomposition. The transformation matrix is ​​then applied to the point clouds of each view, merged into a unified coordinate system, and stitched together.

[0075] 5. Reinforcement model registration (S5)

[0076] Align the extracted rebar point cloud with the pre-stored standard model:

[0077] Construct a local coordinate system: with the background plane normal vector as the Z-axis, the plane projection point of the centroid of the steel reinforcement point cloud as the origin, and the projection direction of the centroid pointing to the first skeleton point as the X-axis, construct a right-handed local coordinate system.

[0078] Global rotation search: Transform the standard model skeleton point set to a local coordinate system. Iterate through all rotation angles from 0° to 359° with a step size of 1°, including cases involving a 180° rotation around the X-axis. Calculate the average distance from the measured skeleton point to the nearest model point in each case, and select the transformation matrix corresponding to the minimum average distance as the final registration result to determine the 3D pose of the reinforcing bars.

[0079] 6. Reinforcing bar dimension inspection (S6)

[0080] In the registered coordinate system, the corresponding detection strategy is executed according to the shape type of the reinforcing bar:

[0081] Straight reinforcing bars: Use the L1 norm to fit a straight line in space, take the first and last two points of the effective point cloud to calculate the distance, and obtain the length of the reinforcing bar.

[0082] Rectangular frame reinforcement: Classify the skeleton point set to four model reference lines, extract the farthest endpoint pair to obtain the four vertices of the rectangle, calculate the distance of each side, and determine the starting end by statistical point cloud density.

[0083] L / U-shaped bent steel bars: straight line fitting is performed on each segment, and the spatial intersection of two adjacent straight lines is calculated as the bend vertex. The length of each segment is calculated with the bend vertex as the boundary.

[0084] Curved steel bars: Using the centroid of the registered skeleton as the geometric center, calculate the distance between two adjacent points and sum them up to obtain the length of the curve.

[0085] 7. Output Results (S7)

[0086] The calculated three-dimensional pose of the reinforcing bar (position coordinates [X,Y,Z] and attitude angles)

[0087] The [α,β,γ] and dimensional parameters (length, angle, etc.) are packaged and output to the robot control system to guide the robotic arm to perform grasping, sorting or processing operations.

[0088] Example 2: Collaborative operation scenario based on AGV and robotic arm

[0089] In this embodiment, the present invention is applied to a mobile robot operation scenario.

[0090] Hardware configuration: A 3D area scan camera is mounted at the end of a six-axis robotic arm, which is mounted on an AGV (Automated Guided Vehicle).

[0091] Collaborative process:

[0092] The AGV moves to the area of ​​the steel bars to be inspected.

[0093] The robotic arm moves to the photo-taking position according to the preset path (if the steel bar is too long, the robotic arm moves to multiple positions to collect data from multiple perspectives).

[0094] Perform steps S1 to S6 in Example 1 above.

[0095] The system calculates the precise gripping posture of the reinforcing bar.

[0096] The robot control system receives position and pose information and controls the robotic arm to precisely grasp steel bars and complete the loading or binding tasks.

[0097] In the tests of the above embodiments, the single detection cycle was consistently between 5 and 10 seconds. For rebars exceeding 3.5 meters in length, the splicing accuracy was controlled within 2 mm through multi-view splicing. This method can accurately identify various types of rebars, including straight, L-shaped, U-shaped, and curved rebars, solving the problems of traditional 2D vision's inability to detect three-dimensional dimensions and the low efficiency of manual measurement.

[0098] The applicant further declares that while the above embodiments illustrate the implementation method and apparatus structure of the present invention, the present invention is not limited to the above-described embodiments, meaning that the present invention must rely on the above methods and structures to be implemented. Those skilled in the art should understand that any improvements to the present invention, equivalent substitutions for the selected implementation methods, additions to steps, and selections of specific methods all fall within the protection and disclosure scope of the present invention.

[0099] This invention is not limited to the above-described embodiments. All methods that employ similar structures and approaches to achieve the objectives of this invention are within the scope of protection of this invention.

Claims

1. A method for precise detection of multiple types of reinforcing bars using 3D vision and robotic collaboration, characterized in that, Includes the following steps: S1: 3D point cloud acquisition, using a 3D area array camera to acquire the original point cloud data and corresponding 2D texture image of the area of ​​the steel bar to be inspected; S2: Background plane detection: Based on the 2D texture image and the original point cloud data, determine the workbench plane where the reinforcing bars are placed and construct the background plane reference. S3: Rebar point cloud segmentation: Based on the background plane reference, extract independent point cloud instances of each rebar from the original point cloud data; S4: Multi-view point cloud stitching. If the length of the steel bar exceeds the field of view of a single camera, the point cloud is registered and stitched by acquiring data from multiple perspectives and using preset markers to generate a complete steel bar point cloud. S5: Rebar model registration, which aligns the extracted or spliced ​​rebar point cloud with the pre-stored standard model to determine the three-dimensional pose of the rebar; S6: Rebar size inspection. In the registered local coordinate system, select the corresponding inspection strategy according to the shape type of the rebar and calculate the size parameters of the rebar. S7: Output the results, outputting the three-dimensional pose and dimensional parameters of the reinforcing bar to the robot control system.

2. The method for precise detection of multiple types of reinforcing bars using 3D vision and robotics collaboration according to claim 1, characterized in that, Step S2: Background plane detection specifically includes: (1) Image enhancement: Select adaptive limited contrast histogram equalization (CLAHE) or multi-scale Retinex (MSR) algorithm according to the size of the platform to enhance the grayscale contrast between the platform and the steel reinforcement area; (2) Adaptive threshold segmentation: The OTSU algorithm is used to automatically determine the threshold for the enhanced grayscale image and generate a planar candidate mask; (3) Morphological refinement: Perform closing and dilation operations on the planar candidate mask in sequence to obtain the refined planar region mask; (4) Depth constraint filtering: Combine the Z-axis depth value of the point cloud, retain only the points whose depth is within the preset window of the target platform height, and remove the points with abnormal depth; (5) Least squares plane fitting: The SVD decomposition is used to perform least squares plane fitting on the selected candidate point set of the plane, solve the normal vector and origin of the plane equation, and output the background plane.

3. The method for precise detection of multiple types of reinforcing bars using 3D vision and robotics collaboration according to claim 1, characterized in that, Step S3: Reinforcing bar point cloud segmentation specifically includes: (1) Planar distance extraction: Calculate the distance from each valid 3D point to the background plane, and retain the points whose distance is within the preset range to generate a set of candidate reinforcement points; (2) Projection binary image generation: Map the candidate point set of steel bars back to 2D image coordinates to generate a candidate region map, and perform dilation operation to connect the fracture region; (3) Connected component analysis and instance segmentation: Perform connected component analysis, assign an independent label to each connected component, filter out noisy areas and incomplete edge areas, and each connected component that passes the filter corresponds to an independent steel bar; (4) 3D instance point cloud recovery: Based on the pixel coordinates of the 2D connected domain, extract the corresponding 3D coordinates from the original point cloud array to obtain an independent 3D point cloud set for each steel bar.

4. The method for precise detection of multiple types of reinforcing bars using 3D vision and robotics collaboration as described in claim 1, characterized in that, Step S4: Multi-view point cloud stitching specifically includes: (1) Circular marker detection: In the 2D texture images of multiple cameras, detect the circular calibration plate pre-set on the worktable and obtain the center coordinates of the marker pixels; (2) Marker 3D coordinate extraction: For each detected circular marker, the effective 3D points near its pixel center are averaged to accurately extract the three-dimensional spatial coordinates of the marker; (3) SVD point-to-point registration: Using the 3D point pairs of markers that are visible to each other in adjacent views as input, the optimal rigid body transformation matrix is ​​solved by SVD decomposition, and the point clouds of multiple views are merged into the same coordinate system to complete the stitching.

5. The method for precise detection of multiple types of reinforcing bars using 3D vision and robotics collaboration according to claim 1, characterized in that, Step S5: Reinforcing bar model registration specifically includes: (1) Construction of local coordinate system: With the background plane normal vector as the Z-axis, the plane projection point of the centroid of the steel reinforcement point cloud as the origin, and the projection direction of the centroid pointing to the first skeleton point as the X-axis, a right-handed local coordinate system is constructed. (2) Global rotation search registration: Transform the pre-stored standard model skeleton point set to the local coordinate system, traverse all rotation angles with a preset step size, including the case of rotation around the axis, calculate the average distance from the measured skeleton point to the nearest model point, and select the transformation matrix corresponding to the minimum average distance as the final registration result.

6. The method for precise detection of multiple types of reinforcing bars using 3D vision and robotics collaboration according to claim 1, characterized in that, In step S6, during the rebar size inspection, different inspection strategies are used for different types of rebar: When the reinforcement is a straight line: fit a straight line in space, take the first and last two points of the effective point cloud, calculate the distance to obtain the length; When the reinforcement is a rectangular frame: classify the skeleton point set into four model reference lines, extract the farthest endpoint pair to obtain the four vertices of the rectangle, calculate the distance between each vertex to obtain the length of each side, and determine the starting end by the point cloud density; For L / U-shaped bent steel bars: perform linear fitting on the point cloud of each steel bar segment, calculate the spatial intersection of two adjacent straight lines to obtain the coordinates of the bend vertex, and calculate the length of each segment with the bend vertex as the boundary; When the reinforcement is curved: take the coordinates of the centroid of the registered skeleton as the reference point of the geometric center, calculate the distance between each two adjacent points and sum them to obtain the length of the curve.

7. The method for precise 3D vision and robot collaborative detection of multiple types of reinforcing bars according to any one of claims 1 to 6, characterized in that, The method is applied to robot collaborative operation scenarios: The robot's end effector is equipped with the 3D area array camera, or the camera is carried by a mobile platform (AGV) and moves in coordination with the robotic arm; In step S1, the robot or mobile platform moves to different camera positions according to a preset path to collect data; After step S7, the robot control system performs subsequent grasping, sorting or processing operations based on the output three-dimensional pose and size parameters.

8. The method for precise detection of multiple types of reinforcing bars using 3D vision and robotic collaboration according to claim 2, characterized in that, In the image enhancement step: for long platforms, the CLAHE algorithm is used with parameters set as follows: clipLimit=3.0, GridSize=11×11; for small platforms, the multi-scale Retinex algorithm is used, with Gaussian scale σ∈{15, 80, 250} decomposed in the logarithmic domain.

9. The method for precise detection of multiple types of reinforcing bars using 3D vision and robotics collaboration according to claim 3, characterized in that, In the planar distance extraction step, the preset range is 5~30mm from the background plane; in the connected component analysis step, noise areas with an area of ​​less than 3000 pixels are filtered out, and areas with a horizontal centroid located within 15% of the left and right edges of the image are excluded.

10. The method for precise detection of multiple types of reinforcing bars using 3D vision and robotics collaboration according to claim 4, characterized in that, The parameters of the circular calibration plate are set as follows: area range of 10,000 to 20,000 pixels, circularity of 0.75 to 1.0; the markers in the middle view are grouped left and right according to the horizontal center line of the image to assist in multi-camera stitching.

Images