A method for spatial ring detection and pose measurement in complex scenes based on trifocal vision

Abstract

This invention discloses a method for detecting spatial rings and measuring pose in complex scenes based on trinocular vision, belonging to the field of spatial target recognition and measurement technology. First, the trinocular vision camera system is calibrated to obtain the internal and external calibration parameters of the AB, CA, and BC binocular combinations, as well as the rotation matrix and displacement vector from the camera to the world coordinate system. Then, an adaptive quadrant-based multi-constraint ellipse detection algorithm is used on the trinocular image of the spatial target in the complex scene to obtain the ellipse information of the inner and outer rings of the target when the edge contour is missing. Next, the normal vectors of the inner and outer rings, as well as the ring radius and pose information, are calculated under the binocular combination. Finally, the target pose information is calculated based on the trinocular vision adaptive weighted pose measurement method. This invention achieves feature recognition and pose measurement navigation for spatial targets in complex scenes through the above technical points, improving the reliability and scene adaptability of the method.

Description

Technical Field

[0001] This invention relates to the field of spatial target recognition and measurement technology, and more specifically, to a method for detecting and measuring spatial rings in complex scenes based on trinocular vision. Background Technology

[0002] The rapid development of space science has led to increasingly complex space activities, and the safety of the space environment has gradually gained importance among countries worldwide. Space activities such as space debris removal, spacecraft refueling, and on-orbit repair and recovery of failed spacecraft have become the focus of technological development both domestically and internationally. Identifying key features and measuring the pose of space targets is a prerequisite for ensuring the success of space missions. Key components of spacecraft are typically circular or near-circular, such as the docking ring between the spacecraft and the launch vehicle and engine nozzles. These features are projected as ellipses on the camera's imaging plane, and their accurate and rapid extraction is crucial for pose measurement. However, factors such as complex space lighting conditions, spacecraft motion blur, and component occlusion can cause incomplete outlines of the target's circular ring. Traditional ellipse detection algorithms and target pose measurement methods often face challenges such as edge breakage, pseudo-ellipse interference, and insufficient fitting accuracy, severely restricting the robustness and real-time performance of pose measurement. For example, the invention "A Binocular Visual Pose Estimation Method Based on Circular Features" (application number 202011356849.4, authorization number CN112381880 B) provides a binocular visual pose estimation method based on circular features. This method obtains the elliptical contour line projected onto the plane by the edge detection and ellipse matching of the target object. It performs feature acquisition and pose feature based on the spatial circle, but does not fully consider the complexity of the scene. Summary of the Invention

[0003] To address the above technical problems, this invention provides a method for detecting and measuring the pose of circular objects in complex scenes based on trinocular vision, which is used to solve the needs of feature recognition and pose measurement navigation when the outline of a circular object in a complex scene is missing.

[0004] To achieve the above technical objectives, the present invention adopts the following specific technical solution: a method for detecting and measuring spatial rings in complex scenes based on trinocular vision, comprising the following steps:

[0005] Step 1: Perform joint calibration on the tri-lens visible light system composed of cameras A, B, and C to obtain the intrinsic and extrinsic parameters, distortion coefficients, rotation matrices and displacement vectors from the coordinate system of each combination to the world coordinate system for the three binocular combinations AB, CA, and BC.

[0006] Step 2: Use the calibration results to correct the distortion of the trinocular image and execute the multi-constraint ellipse detection algorithm based on adaptive quadrant to extract the inner and outer ellipse information of the target spatial ring in the case of missing edges;

[0007] Step 3: Based on the inner and outer ellipse information, calculate the normal vector, radius, and relative pose of the annulus under three binocular combinations: AB, CA, and BC, respectively, to obtain six candidate poses;

[0008] Step 4: Based on the prior size of the spatial ring, the six candidate poses are adaptively weighted and fused to output the final target pose information.

[0009] The present invention has the following beneficial effects:

[0010] This invention provides a multi-constraint ellipse detection algorithm based on adaptive quadrants to address the problem of missing elliptical contours in complex scenes. It adaptively adjusts the number of arc segment quadrants based on the elliptical arc segment characteristics, and performs arc segment matching based on arc segment pre-sorting combined with multiple constraints such as quadrants and relative positions. Subsequently, through coarse fitting, fine fitting, and candidate ellipse scoring mechanisms, it extracts the optimal inner and outer ring elliptical features of the target, improving accuracy, robustness, and computational efficiency. For the adaptive weighted pose measurement method provided by a trinocular vision camera system, it reduces the impact of calibration and single-combination accuracy on the overall result, improving pose measurement accuracy. The purpose of this invention is to provide a trinocular vision-based method for detecting and measuring spatial rings in complex scenes, addressing the feature recognition and pose measurement navigation needs when the contours of spatial target rings are missing in complex scenes. Attached Figure Description

[0011] Figure 1 This is a schematic diagram of the process of the complex scene spatial ring detection and pose measurement method based on trinocular vision of the present invention;

[0012] Figure 2 This is a schematic diagram of the multi-constraint ellipse detection algorithm based on adaptive quadrants provided by the present invention for the problem of missing elliptical contours in complex scenes.

[0013] Figure 3 This is a schematic diagram illustrating the effect of detecting circular rings in complex scenes according to the present invention.

[0014] Figure 4 This is a schematic diagram of the adaptive weighted pose measurement method for a trinocular vision camera system provided by the present invention. Detailed Implementation

[0015] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other. To achieve the above objectives, this invention adopts the following technical solution.

[0016] Example:

[0017] refer to Figure 1 A method for detecting and gaiting circular objects in complex scenes based on trinocular vision includes the following steps:

[0018] Step 1: Calibrate a trinocular vision camera system consisting of visible light cameras A, B, and C of the same model and specifications. Obtain the calibration parameters and distortion coefficients of the AB, CA, and BC binocular combination, and obtain the rotation matrix and translation vector from the binocular combination coordinate system to the world coordinate system. The specific method is as follows:

[0019] Step 1.1: Simultaneously acquire images of the checkerboard calibration board of the trinocular vision camera system, and use the Zhang Youzheng calibration method to perform single-target calibration of camera A, camera B, and camera C, and dual-target calibration of AB, CA, and BC respectively, to obtain the system calibration parameters and distortion coefficients;

[0020] Step 1.2: Set the world coordinate system of the trinocular vision camera system, and use a total station to measure the three-dimensional coordinates of the 16 markers on the chessboard in the world coordinate system;

[0021] Step 1.3: Extract the two-dimensional coordinates of the 16 marker points of the chessboard grid on the imaging planes of cameras A, B, and C, and calculate their three-dimensional coordinates in the combined binocular coordinate system of cameras AB, CA, and BC.

[0022] Step 1.4: Using the 3D coordinates of the 16 marker points on the checkerboard pattern in the world coordinate system and the combined coordinate systems of cameras AB, CA, and BC, calculate the rotation matrix between the world coordinate system and each combined coordinate system. , , and displacement vector , , The three-dimensional coordinates of point P in space, under the combined view of cameras AB, CA, and BC, are: , and Its three-dimensional coordinates in the world coordinate system are:

[0023] ;

[0024] Step 2: Reference Figure 2 and Figure 3 This method acquires trinocular images of spatial targets in complex scenes, corrects image distortion based on camera calibration parameters and distortion coefficients, and employs an adaptive quadrant-based multi-constraint ellipse detection algorithm to obtain the ellipse information of the inner and outer rings of the target when the edge contour is missing. The adaptive quadrant-based multi-constraint ellipse detection algorithm is as follows:

[0025] Step 2.1: Perform Gaussian filtering preprocessing and edge arc detection on the acquired trinocular image of the spatial target in the complex scene. Use adaptive quadrant division to determine the number and range of quadrants. Use bitmask to record the quadrants covered by the arcs. Extract the arcs in the image. Connect the arcs that meet the conditions of continuity and concavity / convexity to form arc groups. All arc groups constitute arc combinations.

[0026] Step 2.2: Pre-sort the arc segments in the arc segment combination, and perform arc segment matching by combining multiple constraints such as quadrant and relative position to form a candidate arc set for coarse fitting of the ellipse;

[0027] Step 2.3: Perform least squares fitting on all points in the candidate arc subset to obtain the coarse fitting result of the ellipse corresponding to the candidate arc subset:

[0028] ;

[0029] in, Represents the semi-major axis of the ellipse. Represents the minor semi-axis of the ellipse. This represents the points on the selected subset of candidate arcs. Let be the center of the ellipse. The angle of inclination of the ellipse;

[0030] Step 2.4: Use the adaptive weighted iterative ellipse fitting based on the gradient of the edge points to obtain a set of candidate ellipses;

[0031] Step 2.5: Based on the geometric characteristics of the spatial annulus, establish a candidate ellipse scoring mechanism to filter the ellipse information of the inner and outer rings of the target when the edge contour is missing;

[0032] Step 3: Based on the elliptical information of the inner and outer rings of targets A, B, and C in Step 2, and according to the dual-target positioning parameters and the rotation matrix from the binocular combined coordinate system to the world coordinate system... , , and displacement vector , , Calculate the inner loop normal vector under the combined A and B cameras. Inner ring radius and relative position and attitude outer ring normal vector Outer ring radius and relative position and attitude Inner loop normal vector under camera CA binocular combination Inner ring radius and relative position and attitude outer ring normal vector Outer ring radius and relative position and attitude Inner loop normal vector under the BC binocular combination of cameras Inner ring radius and relative position and attitude outer ring normal vector Outer ring radius and relative position and attitude A total of 6 sets of binocular combination results;

[0033] Step 4: Reference Figure 4 By combining prior information on the dimensions of the docking ring of the space target, the pose results of the six binocular combinations are adaptively weighted to obtain the target pose information.

[0034] In this implementation, step 2.1, the method for extracting arc segments and dividing them into quadrants from the acquired trinocular image of the spatial target in the complex scene to form arc segment groups, is as follows:

[0035] Step 2.1.1: Calculate the gradient direction of all straight line segments that make up the arc. Convert it into a two-dimensional gradient vector As input features for subsequent clustering;

[0036] Step 2.1.2: For all extracted two-dimensional gradient vectors Use K-means clustering to define the candidate quadrant set. For each quadrant number Clustering algorithms are performed on all quadrants to obtain the clustering results for each quadrant number, as well as the total squared distance corresponding to each quadrant number. :

[0037] ;

[0038] in, Quadrant quantity This indicates that all data is assigned to the cluster center. gradient direction , This represents the circumferential distance between the gradient direction and the cluster center;

[0039] Step 2.1.3: Use the Bayesian information criterion as an indicator to evaluate cluster quality:

[0040] ;

[0041] Where N is the total number of line segments, and records the number of quadrants that minimize the BIC. Select the optimal number of quadrants and cluster centers;

[0042] Step 2.1.4: Select the optimal number of quadrants The cluster centers are converted back to angular representations, and the quadrant boundary is defined as the midpoint of the angle between two adjacent cluster centers. The angular space is then dynamically divided into... In each quadrant;

[0043] Step 2.1.5: For each arc segment, traverse all the line segments formed by it, calculate the gradient direction and map it to the corresponding quadrant region, and use a bit mask to count the covered quadrants and their number.

[0044] Step 2.1.6: Connect the arc segments that satisfy the continuity and concavity / convexity conditions to form an arc segment group. Continuity means that the Euclidean distance between the head and tail vertices of two adjacent arc segments should be less than a certain threshold; concavity / convexity means that the included angle between arc segment regions is less than a threshold and the angle between the horizontal angle of the pixel and the support region of the candidate line segment is less than a threshold.

[0045] In this implementation, step 2.2 involves pre-sorting the arc segment combinations and performing arc segment matching based on multiple constraints such as quadrants and relative positions to form a candidate arc set. Specifically:

[0046] Step 2.2.1: Calculate the gradient difference of each sub-arc group in the arc combination; use the gradient difference between its first and last arcs as the coverage angle of the sub-arc group to obtain the coverage angle of each sub-arc group in the arc combination.

[0047] Step 2.2.2: Select arc segments with a coverage angle greater than a preset angle threshold as a subset of the candidate arc set;

[0048] Step 2.2.3: Combine all arc segments in the arc segment combination into pairs, and filter out the arc segment combinations that simultaneously satisfy the following conditions: the sum of the coverage angles after combination is greater than the preset angle threshold, they come from different quadrants, and their relative positions are greater than the preset positions. These are also used as a subset of the candidate arc set.

[0049] In this implementation, step 2.4 uses adaptive weighted iterative ellipse fitting with edge point gradients to obtain a set of candidate ellipses, specifically as follows:

[0050] Step 2.4.1: Count the number of edge points on the ellipse of the arc segments of the candidate arc subset participating in the coarse fitting. Select the candidate arc subset whose number is close to the length of the arc segment itself or close to the total number of potential edge points in the local area as the effective arc segments of the ellipse, and as the candidate arc subset of the set of effective arc segments of the ellipse.

[0051] Step 2.4.2: Search for the remaining arc segments near the ellipse containing the above effective arc segments, and calculate the minimum weighted error square value of the arc segments. The specific method is as follows:

[0052] For each edge point of the remaining arc segment, calculate its weight. Define this edge point normalized gradient magnitude The cosine of the difference between its gradient direction and the theoretical gradient direction of the ellipse The product of:

[0053] ;

[0054] in, It is the normalized difference between the actual gradient direction at the edge point and the theoretical gradient direction on the ellipse. It is an edge point The theoretical gradient direction, It is the actual gradient direction at that point. It is a preset threshold, taking into account the effects of light and noise, generally Set to 10°;

[0055] The minimum weighted squared error value for each arc segment is calculated and defined as:

[0056] ;

[0057] in, Represents edge points The weights are given by N, where N represents the number of candidate arc edge points.

[0058] Step 2.4.3: When the weighted squared error of the remaining arc segments is less than a preset threshold, it is taken as a candidate arc subset of the set of effective arc segments of the ellipse;

[0059] Step 2.4.4: Perform least squares fitting on all points in the subset of effective arc segments of the ellipse to obtain the candidate ellipse set.

[0060] In this implementation, step 2.5, the candidate ellipse scoring mechanism and the optimal inner and outer ring ellipse information filtering method, are specifically as follows:

[0061] Step 2.5.1: The candidate ellipse scoring mechanism function is expressed as:

[0062] ;

[0063] in, This represents the ratio of the total number of edge points in the set of effective arc segments of the ellipse to the circumference of the ellipse. This represents the ratio of the central angle corresponding to the largest continuous arc segment that can be formed from all edge points of the effective arc segment set to the total angle covered by it on the ellipse. This represents the proportion of the angles covered by all edge points of the effective arc segment set on the ellipse. , , This is the rating coefficient;

[0064] Step 2.5.2: Perform pseudo-sorting on the candidate ellipse set according to the above evaluation, remove the low-scoring candidate ellipse subset fitted by the same arc segment, and obtain the information of the inner and outer ring ellipses with the highest scores.

[0065] In this implementation, step 4, the adaptive weighted pose measurement method, is specifically as follows:

[0066] Step 4.1: Calculate the position weights of the stereo combination of cameras AB, CA, and BC. and pose weights , i=1~6. When the difference between the calculated radius and the true radius is greater than a preset threshold, the position weight of this combination is 0; when the angle between the circular normal vector and the mean normal vector is greater than a preset threshold, the attitude weight of this combination is 0. Specifically:

[0067] ;

[0068] ;

[0069] in, This represents the difference between the calculated radius of the annulus and the actual radius. The threshold value for the preset radius difference; This represents the difference between the normal vector and the mean of the annulus. A threshold value is preset for the difference between the angles of the normal vectors; To avoid extremely small constants with a denominator of 0; , These are the actual results for the radius and normal vector of the annulus, respectively.

[0070] Step 4.2: Based on the adaptive weighted results of the six sets of binocular combined poses, the target pose information is obtained, represented as:

[0071] , .

[0072] This specific embodiment is merely an explanation of the present invention and is not intended to limit the invention. After reading this specification, those skilled in the art can make modifications to this embodiment without contributing any inventive step, but such modifications are protected by patent law as long as they are within the scope of the claims of the present invention.

Claims

1. A method for detecting and measuring the pose of spatial rings in complex scenes based on trinocular vision, characterized in that, include: Step 1: Perform joint calibration on the tri-lens visible light system composed of cameras A, B, and C to obtain the intrinsic and extrinsic parameters, distortion coefficients, rotation matrices and displacement vectors from the coordinate system of each combination to the world coordinate system for the three binocular combinations AB, CA, and BC. Step 2: Use the calibration results to correct the distortion of the trinocular image and execute the multi-constraint ellipse detection algorithm based on adaptive quadrant to extract the inner and outer ellipse information of the target spatial ring in the case of missing edges; Step 3: Based on the inner and outer ellipse information, calculate the normal vector, radius, and relative pose of the annulus under three binocular combinations: AB, CA, and BC, respectively, to obtain six candidate poses; Step 4: Based on the prior size of the spatial ring, the six candidate poses are adaptively weighted and fused to output the final target pose information.

2. The method for detecting and measuring spatial rings in complex scenes based on trinocular vision according to claim 1, characterized in that, in, Step 1 specifically includes: Step 1.1: Simultaneously acquire images of the checkerboard calibration board of the trinocular vision camera system, and use the Zhang Youzheng calibration method to perform single-target calibration of camera A, camera B, and camera C, and dual-target calibration of AB, CA, and BC respectively, to obtain the system calibration parameters and distortion coefficients; Step 1.2: Set the world coordinate system of the trinocular vision camera system, and use a total station to measure the three-dimensional coordinates of the 16 markers on the chessboard in the world coordinate system; Step 1.3: Extract the two-dimensional coordinates of several checkerboard markers on the imaging planes of cameras A, B, and C, and calculate their three-dimensional coordinates in the combined binocular coordinate system of cameras AB, CA, and BC; Step 1.4: Using the 3D coordinates of several marker points on the checkerboard pattern in the world coordinate system and the combined coordinate systems of cameras AB, CA, and BC, calculate the rotation matrix between the world coordinate system and each combined coordinate system. , , and displacement vector , , .

3. The method for detecting and measuring spatial rings in complex scenes based on trinocular vision according to claim 1, characterized in that, in, Step 2 specifically includes: Step 2.1: Perform Gaussian filtering preprocessing and edge arc detection on the acquired triocular image. Use adaptive quadrant division to determine the number and range of quadrants. Use bitmask to record the quadrants covered by the arcs. Extract the arcs in the image. Connect the arcs that meet the conditions of continuity and concavity / convexity to form arc groups. All arc groups constitute arc combinations. Step 2.2: Pre-sort the arc segments in the arc segment combination, and perform arc segment matching by combining quadrant and relative position multiple constraints to form a candidate arc set for coarse fitting of the ellipse; Step 2.3: Perform least squares fitting on all points in the candidate arc subset to obtain the coarse fitting result of the ellipse corresponding to the candidate arc subset: ； in, Represents the semi-major axis of the ellipse. Represents the minor semi-axis of the ellipse. This represents the points on the selected subset of candidate arcs. Let be the center of the ellipse. The angle of inclination of the ellipse; Step 2.4: Use the adaptive weighted iterative ellipse fitting based on the gradient of the edge points to obtain a set of candidate ellipses; Step 2.5: Based on the geometric characteristics of the spatial annulus, establish a candidate ellipse scoring mechanism to filter the ellipse information of the inner and outer rings of the target when the edge contour is missing.

4. The method for detecting and measuring spatial rings in complex scenes based on trinocular vision according to claim 1, characterized in that, Step 3 specifically involves: based on the ellipse information of the inner and outer loops, and according to the dual-camera positioning parameters and the rotation matrix and displacement vector from the binocular combined coordinate system to the world coordinate system, calculating the inner loop normal vector under the camera AB binocular combination. Inner ring radius and relative position and attitude outer ring normal vector Outer ring radius and relative position and attitude Inner loop normal vector under camera CA binocular combination Inner ring radius and relative position and attitude outer ring normal vector Outer ring radius and relative position and attitude Inner loop normal vector under the BC binocular combination of cameras Inner ring radius and relative position and attitude outer ring normal vector Outer ring radius and relative position and attitude .

5. The method for detecting and measuring spatial rings in complex scenes based on trinocular vision according to claim 3, characterized in that, Step 2.1 includes: Step 2.1.1: Calculate the gradient direction of all straight line segments that make up the arc. Convert it into a two-dimensional gradient vector As input features for subsequent clustering; Step 2.1.2: For all extracted two-dimensional gradient vectors Use K-means clustering to define the candidate quadrant set. For each quadrant number Clustering algorithms are performed on all quadrants to obtain the clustering results for each quadrant number, as well as the total squared distance corresponding to each quadrant number. : ； in, Quadrant quantity This indicates that all data is assigned to the cluster center. gradient direction , This represents the circumferential distance between the gradient direction and the cluster center; Step 2.1.3: Use the Bayesian information criterion as an indicator to evaluate cluster quality: ； Where N is the total number of line segments, and records the number of quadrants that minimize the BIC. Select the optimal number of quadrants and cluster centers; Step 2.1.4: Select the optimal number of quadrants The cluster centers are converted back to angular representations, and the quadrant boundary is defined as the midpoint of the angle between two adjacent cluster centers. The angular space is then dynamically divided into... In each quadrant; Step 2.1.5: For each arc segment, traverse all the line segments formed by it, calculate the gradient direction and map it to the corresponding quadrant region, and use a bit mask to count the covered quadrants and their number. Step 2.1.6: Connect the arc segments that meet the conditions of continuity and concavity / convexity to form an arc segment group. Continuity means that the Euclidean distance between the head vertex and the tail vertex of two adjacent arc segments should be less than a certain threshold. Concavity / convexity means that the included angle between arc segment regions is less than a threshold and the angle between the horizontal angle of the pixel and the angle of the candidate line segment support region is less than a threshold.

6. The method for detecting and measuring spatial rings in complex scenes based on trinocular vision according to claim 3, characterized in that, Step 2.2 includes: Step 2.2.1: Calculate the gradient difference of each sub-arc group in the arc combination; use the gradient difference between its first and last arcs as the coverage angle of the sub-arc group to obtain the coverage angle of each sub-arc group in the arc combination. Step 2.2.2: Select arc segments with a coverage angle greater than a preset angle threshold as a subset of the candidate arc set; Step 2.2.3: Combine all arc segments in the arc segment combination into pairs, and filter out the arc segment combinations that simultaneously satisfy the following conditions: the sum of the coverage angles after combination is greater than the preset angle threshold, they come from different quadrants, and their relative positions are greater than the preset positions. These are also used as a subset of the candidate arc set.

7. The method for detecting and measuring spatial rings in complex scenes based on trinocular vision according to claim 3, characterized in that, Step 2.4 includes: Step 2.4.1: Count the number of edge points on the ellipse of the arc segments of the candidate arc subset participating in the coarse fitting. Select the candidate arc subset whose number is close to the length of the arc segment itself or close to the total number of potential edge points in the local area as the effective arc segments of the ellipse, and as the candidate arc subset of the set of effective arc segments of the ellipse. Step 2.4.2: Search for the remaining arc segments near the ellipse containing the above effective arc segments, and calculate the minimum weighted squared error value of the arc segments; Step 2.4.3: When the weighted squared error of the remaining arc segments is less than a preset threshold, it is taken as a candidate arc subset of the set of effective arc segments of the ellipse; Step 2.4.4: Perform least squares fitting on all points in the subset of effective arc segments of the ellipse to obtain the candidate ellipse set.

8. The method for detecting and measuring spatial rings in complex scenes based on trinocular vision according to claim 7, characterized in that, The specific method for step 2.4.2 is as follows; For each edge point of the remaining arc segment, calculate its weight. Define this edge point normalized gradient magnitude The cosine of the difference between its gradient direction and the theoretical gradient direction of the ellipse The product of: ； in, It is the normalized difference between the actual gradient direction at the edge point and the theoretical gradient direction on the ellipse. It is an edge point The theoretical gradient direction, It is the actual gradient direction at that point. It is a preset threshold; The minimum weighted squared error value for each arc segment is calculated and defined as: ； in, Represents edge points The weights are given by N, which represents the number of candidate arc edge points.

9. The method for detecting and measuring spatial rings in complex scenes based on trinocular vision according to claim 3, characterized in that, Step 2.5 includes: Step 2.5.1: The candidate ellipse scoring mechanism function is expressed as: ； in, This represents the ratio of the total number of edge points in the set of effective arc segments of the ellipse to the circumference of the ellipse. This represents the ratio of the central angle corresponding to the largest continuous arc segment that can be formed from all edge points of the effective arc segment set to the total angle covered by it on the ellipse. This represents the proportion of the angles covered by all edge points of the effective arc segment set on the ellipse. , , This is the rating coefficient; Step 2.5.2: Perform pseudo-sorting on the candidate ellipse set according to the above evaluation, remove the low-scoring candidate ellipse subset fitted by the same arc segment, and obtain the information of the inner and outer ring ellipses with the highest scores.

10. The method for detecting and measuring spatial rings in complex scenes based on trinocular vision according to claim 1, characterized in that, Step 4 includes: Step 4.1: Calculate the position weights of the stereo combination of cameras AB, CA, and BC. and pose weights For i=1~6, when the difference between the calculated radius and the true radius is greater than a preset threshold, the position weight is 0; when the angle between the circular normal vector and the mean normal vector is greater than a preset threshold, the attitude weight is 0. Specifically: ； ； in, This represents the difference between the calculated radius of the annulus and the actual radius. The threshold value for the preset radius difference; This represents the difference between the normal vector and the mean of the annulus. A threshold value is preset for the difference between the angles of the normal vectors; To avoid extremely small constants with a denominator of 0; , These are the actual results for the radius and normal vector of the annulus, respectively. Step 4.2: Based on the adaptive weighted results of the six sets of binocular combined poses, the target pose information is obtained, represented as: ， 。

Images