Fast positioning method for multi-shaped targets based on target geometric constraints

Abstract

The present invention discloses a fast positioning method for multi-shaped targets based on target geometric constraints, comprising the following steps: Step 1, collect an image sequence S and obtain the shape type Type of the target; Step 2, select corresponding image segmentation and feature point extraction methods to obtain the coordinates of the target feature points in the pixel coordinate system; Step 3, establish a constraint equation set according to the camera projection equation and the coordinates of the target feature points in the pixel coordinate system, in combination with the target geometric constraint information; Step 4, solve the constraint equation set by an improved Newton iteration method to obtain the depth values of the feature points; Step 5, fuse the camera poses obtained by monocular SLAM to obtain the coordinates of the feature points in the world coordinate system. The present invention only needs a monocular image and the geometric information of the target to complete the three-dimensional space positioning of targets of various shapes, with relatively low hardware costs and computational costs required, a simple system, small computational amount, and can achieve a relatively accurate target positioning effect.

Description

Technical Field

[0001] The present invention belongs to the technical field of computer vision processing, and particularly relates to a method for quickly locating multi-shaped targets based on target geometric constraints. Background Art

[0002] Vision sensors have the characteristics of low cost and rich image sequence information obtained. Based on image information, applications such as environmental perception, visual positioning, and visual navigation can be realized. Therefore, vision sensors are widely used in fields such as robots and autonomous driving. The working principle of a vision sensor: The light reflected by an object in the three-dimensional world passes through the camera lens and is focused on the camera imaging plane, which realizes the mapping of the coordinate points in the world coordinate system in the three-dimensional space to the image coordinate system on the two-dimensional imaging plane.

[0003] As an important tool for environmental detection and exploration of unknown areas, the mobile robot system is developing towards the trends of intelligence, autonomy, miniaturization, and low cost. When the robot system is operating, it often faces specific objects, which requires the mobile robot system to be able to achieve spatial positioning of the working object. Vision-based target positioning refers to relying on the image information collected by a vision sensor to restore the three-dimensional spatial coordinate information of the target. In essence, it is the remapping of the pixel coordinate system of the image to the world coordinate system. Compared with the world coordinate system, the pixel coordinate system lacks one dimension. Obviously, the remapping process cannot be completed through a single image. Existing technologies include using binocular vision, 3D structured light cameras with speckle coding, etc. to directly obtain the depth information of the image through hardware devices. This method has a high cost and it is difficult to miniaturize the hardware; or using monocular vision based on the triangulation principle, and realizing the method of recovering the three-dimensional structure from motion through a large number of associated images. Its essence is still the simulation of stereo vision, which requires estimating the baseline distance and then completing the depth estimation of the image. The related methods have low accuracy; and the image depth estimation method based on deep learning. At present, the research results in this direction have a large amount of calculation, and the estimation accuracy and real-time performance are still difficult to meet the requirements of target navigation and positioning on mobile platforms.

[0004] Generally, a monocular vision system cannot directly recover the depth information of a target from a two-dimensional image. At present, the vision-based depth estimation technologies have disadvantages such as high cost, difficulty in miniaturization, large amount of calculation, and low accuracy, and are difficult to meet the applications on miniaturized and low-cost mobile robot platforms. Summary of the Invention

[0005] The purpose of the present invention is to overcome the deficiencies of the prior art, and provide a method for quickly locating multi-shaped targets based on target geometric constraints that can complete the three-dimensional spatial positioning of multi-shaped targets only with a monocular image and the geometric information of the target, with relatively low required hardware cost and calculation cost, a simple system, a small amount of calculation, and can achieve a relatively accurate target positioning effect.

[0006] The object of the present invention is achieved by the following technical solutions: A multi-shape object rapid positioning method based on target geometric constraints, comprising the following steps:

[0007] Step 1: Use the vision sensor carried by the mobile robot platform to collect images, arrange them in chronological order to form an image sequence S, and obtain the shape type Type of the object in the image through the target detection algorithm;

[0008] Step 2: According to the shape type Type of the object, select the corresponding image segmentation and feature point extraction method to obtain the coordinates P of the target feature points in the pixel coordinate system uv i = [u i , v i , 1] T , i = 1, 2,..., n, where n is the number of feature points;

[0009] Step 3: According to the camera projection equation and the coordinates P of the target feature points in the pixel coordinate system uv i = [u i , v i , 1] T , combined with the target geometric constraint information to establish a constraint equation system;

[0010] Step 4: Solve the constraint equation system through an improved Newton iteration method to obtain the depth values of the feature points;

[0011] Step 5: Fuse the camera pose [R|t] obtained by monocular SLAM to obtain the coordinates P of the feature points in the world coordinate system i = [X i , Y i , Z i T , i = 1, 2,..., n, to achieve the rapid positioning of multi-shape objects, and the mathematical description is as follows:

[0012] P w = RP i + t

[0013] R and t are the rotation matrix and translation vector describing the camera pose change respectively.

[0014] Furthermore, the shape type Type of the object includes parallelogram, circle or triangle.

[0015] Furthermore, the specific implementation method of the said Step 2 is:

[0016] Step 21: Perform threshold processing on the detected image to obtain a binary image;

[0017] Step 22: Obtain the edge information of the target in the image through the adaptive Canny edge extraction algorithm;

[0018] Step 23: Use the Hough line transform to extract the line features in the parallelogram or triangle edge image, and use the Hough circle transform to extract the arc features in the circular edge image;

[0019] Step 24: For the line features, extract the feature corner points through Harris corner detection; for the arc features, randomly select two points on the arc edge, and respectively take the intersection points of the lines connecting the two points and the center of the circle with the opposite arc as the other two feature corner points. If the connection line cannot generate an intersection point with the known arc segment, reselect the points;

[0020] Step 25: Eliminate the abnormal feature corner points according to the target category Type and the known target scale information.

[0021] Furthermore, the specific implementation method of the said Step 3 is as follows:

[0022] A. Use the distance constraint and parallel constraint of the parallelogram to establish the geometric constraint equation set of the parallelogram image. The specific method is: Take the four corner points of the parallelogram as the feature points to describe the target spatial position. The coordinates of the four points in the world coordinate system are P i =[X i ,Y i ,Z i T ,i = 1, 2, 3, 4; The coordinates in the camera coordinate system are P w i=[P wxi ,P wyi ,P wzi T ; At the same time, in the normalized plane, the normalized projection point coordinates of the target are expressed as P c i=[p cxi ,p cyi ,1] T ; The coordinates of the target in the pixel coordinate system are P uv i=[u i ,v i ,1] T ; d ij represents the distance between the connection lines of corner points i and j;

[0023] Establish 6 distance constraint equations:

[0024]

[0025] i,j = 1, 2, 3, 4 and i≠j

[0026] According to the condition that the opposite sides of the parallelogram are parallel ​​Establish 4 parallel constraint equations:

[0027]

[0028]

[0029]

[0030]

[0031] For the parallelogram target, establish a total of 10 equations including distance constraints and parallel constraints; the four unknowns in the equations are the Z-axis components p of the coordinates of the four corner points of the parallelogram door frame wzi ; Select any two equations from the distance constraints and add the two equations describing the parallelism of a pair of opposite sides to form a well-posed system of equations, and solve for the depth values of the pixel coordinates of the four corner points of the parallelogram target;

[0032] B. The projection of a circle is a circle or an ellipse, and the projection of the center of the circle is still the center of symmetry of the projected figure; the line connecting the intersection points of the line passing through this center and the image edge is the diameter; take the intersection points of two lines passing through the center of the circle and the image edge to obtain four points p c1 , p c2 , p c3 , p c4 , and their coordinates are P c i = [p cxi , p cyi , 1] T , i = 1, 2, 3, 4, and the coordinates of the four points in the world coordinate system are P i = [X i , Y i , Z i T , and the coordinates in the camera coordinate system are P w i = [P wxi , P wyi , P wzi T , and the coordinates in the pixel coordinate system are P uv i = [u i , v i , 1] T ; The four points can form a parallelogram when connected in sequence; for the circular target, only use the diagonal distance as the distance constraint and establish the following distance constraint equations:

[0033]

[0034] i = 1, j = 3 or i = 2, j = 4

[0035] Then, four constraint equations are established in a manner consistent with the parallel condition of the opposite sides of the parallelogram in A;

[0036] For a circular target, there are a total of six equations including distance constraints and parallel constraints. Select two equations of the distance constraints and any set of two equations describing the parallelism of opposite sides to form a well-posed system of equations, and calculate the depth values of the pixel coordinates of the four feature points of the circular target;

[0037] C. For a triangular target, the side length of the triangle is known as d ij ; First, three equations are established according to the distance constraint:

[0038]

[0039] i, j = 1, 2, 3 and i ≠ j

[0040] If the triangle is an equilateral triangle, according to the properties of an equilateral triangle, the line connecting any vertex to the midpoint of the opposite side is perpendicular to the opposite side; the mathematical expression of the vector perpendicular relationship in the Cartesian coordinate system is:

[0041]

[0042] For an equilateral triangular target, there are a total of four equations including distance constraints and perpendicular constraints. Select any two equations of the distance constraints and the equation describing the vector perpendicular relationship to form a well-posed system of equations, and calculate the depth values of the pixel coordinates of the three feature points of the equilateral triangular target;

[0043] If the triangle is a non-equilateral triangle, the depth values of the three feature points are calculated by the three equations established through the distance constraint to determine the three-dimensional spatial position of the triangular target.

[0044] Furthermore, the specific implementation method of step 4 is as follows: The input of the improved Newton iteration method is the system of multivariate nonlinear equations constructed in step 3, and the output is the depth value of the feature point; The improved Newton iteration method with fifth-order convergence is adopted, and the iteration equation is:

[0045]

[0046]

[0047]

[0048] X n is an unknown vector composed of the depth values p wzi of the feature points, and Y n , Z n are intermediate quantities in the iteration process;

[0049] Set the termination condition of the iteration algorithm as:

[0050] ||X n+1 -X n ||1 < 0.01 or η ≥ 20

[0051] where η represents the number of iterations, and the maximum number of iterations is 20; substituting the calculated depth value p of the feature point wzi into the camera projection equation to determine the coordinate point P of each feature point in the camera coordinate system w i:

[0052]

[0053] The beneficial effects of the present invention are as follows: The present invention uses the geometric features of multiple-shaped targets to establish a geometric constraint equation set, which has the characteristics of clear physical meaning, simple and fast solution. Only a monocular image and the geometric information of the target are required to complete the three-dimensional space positioning of multiple-shaped targets. The required hardware cost and calculation cost are both low, the system is simple, the calculation amount is small, and a relatively accurate target positioning effect can be achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 is a flowchart of the multi-shaped target rapid positioning method of the present invention;

[0055] Figure 2 is a flowchart of obtaining the coordinates of the target feature point in the pixel coordinate system of the present invention;

[0056] Figure 3 is a schematic diagram of the target geometric constraint of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0057] Definition of three coordinate systems involved in the target positioning process of the present invention:

[0058] 1. World coordinate system: A fixed coordinate system defined during algorithm initialization, used as a reference coordinate system to describe the absolute position of the target in three-dimensional space. In this paper, P = [X, Y, Z] T is used to represent.

[0059] 2. Camera coordinate system: The origin of the camera coordinate system coincides with the camera optical center O, the Z-axis points to the front of the camera, the Y-axis points to the top of the camera, and the X-axis points to the left of the camera. In this paper, P w = [P wx , P wy , P wz T is used to represent. Particularly, in the normalized plane, the normalized projection point coordinates of the target are represented as P c = [p cx , p cy , 1] T . ​

[0060] 3. Pixel coordinate system: The pixel coordinate system is fixed on the camera projection imaging plane. The origin O' is defined at the upper left corner of the image. The u-axis points to the right of the image, and the v-axis points downward. Then the coordinates of the target in the pixel coordinate system are P uv = [u, v, 1] T .

[0061] As Figure 1 shown, a fast multi-shape target localization method based on target geometric constraints of the present invention includes the following steps:

[0062] Step 1: Use the vision sensor carried by the mobile robot platform to collect images, arrange them in chronological order to form an image sequence S, and obtain the shape type Type of the target in the image through the target detection algorithm; the shape type Type of the target includes parallelogram, circle or triangle.

[0063] Step 2: According to the shape type Type of the target, select the corresponding image segmentation and feature point extraction method to obtain the coordinates P uv i = [u i , v i , 1] T , i = 1, 2,..., n, where n is the number of feature points; as Figure 2 shown, the specific implementation method is as follows:

[0064] Step 21: Perform threshold processing on the detected image to obtain a binary image;

[0065] Step 22: Obtain the edge information of the target in the image through the adaptive Canny edge extraction algorithm;

[0066] Step 23: Use the Hough line transform to extract the line features in the parallelogram or triangle edge image, and use the Hough circle transform to extract the arc features in the circle edge image;

[0067] Step 24: For the line features, extract the feature corner points through the Harris corner detection; for the arc features, randomly select two points on the arc edge, and respectively take the intersection points of the lines connecting the two points and the center of the circle with the opposite arc as the other two feature corner points. If the connection line cannot generate an intersection point with the known arc segment, reselect the points;

[0068] Step 25: Eliminate abnormal feature corner points according to the target category Type and the known target scale information:

[0069] Parallelogram: According to the known conditions that the opposite sides of a parallelogram are parallel and the lengths of the opposite sides are equal, determine whether there is a parallel or approximately parallel relationship between the lines formed by the currently extracted corner points, and whether the root mean square error of the lengths of the opposite sides exceeds a preset threshold. Discard the feature corner points and the corresponding images with errors exceeding the threshold, and directly proceed to the calculation of the next image;

[0070] Circle: The same as the method for removing outliers of parallelogram features;

[0071] Triangle: Determine whether the root mean square error of the distances between any two of the three feature points of the triangle exceeds a preset threshold. Discard the feature corner points and the corresponding images with errors exceeding the threshold, and directly proceed to the calculation of the next image.

[0072] Step 3: According to the camera projection equation and the coordinates P of the target feature points in the pixel coordinate system uv i = [u i , v i , 1] T , establish a constraint equation set in combination with the target geometric constraint information;

[0073] Taking the parallelogram target as an example to illustrate the principle of the present invention: In this paper, four corner points of the parallelogram are taken as feature points to describe the target spatial position. The coordinates of the target in the world coordinate system are P i = [X i , Y i , Z i T , i = 1, 2, 3, 4. The coordinates in the camera coordinate system are P w i = [P wxi , P wyi , P wzi T , i = 1, 2, 3, 4. At the same time, in the normalized plane, the normalized projection point coordinates of the target are expressed as P c i = [p cxi , p cyi , 1] T , i = 1, 2, 3, 4. The coordinates of the target in the pixel coordinate system are P uv i = [u i , v i , 1] T , i = 1, 2, 3, 4.

[0074] Assume that the pixel coordinates are scaled by α on the u-axis and by β on the v-axis, and the origin is translated by [c x , c y T . For any coordinate point P i in three-dimensional space, the projection point P w in the camera coordinate systemi and the imaging point P in the pixel coordinate system uv The coordinates of i have the following relationship:

[0075]

[0076] Merge αf into f x and merge βf into f y Then the above formula can be transformed into:

[0077]

[0078] Use subscripts to distinguish the coordinate systems, and at the same time convert the above formula into the form of an augmented matrix as:

[0079]

[0080] The 3×3 matrix in the above formula is called the intrinsic matrix and is represented by the symbol K. The camera intrinsics are fixed and can be obtained through camera calibration for the used monocular camera's intrinsic matrix.

[0081] The camera coordinate system is fixed on the camera and moves with the camera. The coordinate P of the same target in the world coordinate system can be described by a rotation matrix R and a translation vector t i and the coordinate P w i in the camera coordinate system.

[0082] P w = RP i + t

[0083] Substitute to get:

[0084]

[0085] The coordinates in the world coordinate system have been transformed to the camera coordinate system, and the Z-axis can be further normalized to obtain the projection of point P i on the camera normalization plane:

[0086]

[0087] The form of the constraint equation system established according to the target geometric constraint information described in step 3 is as follows: Any pixel point in the two-dimensional image can be expressed in the following form, and there is only one unknown p in the formula wzi . If the true distance relationship between multiple pixel points is known, then the following equations can be established:

[0088]

[0089] When the number of equations that can be established is greater than the number of unknowns, the equations can be solved, that is Obviously, when we know the distance relationships among more than three points, we can establish distance equations pairwise and calculate the depth information of each feature point. Solving the distance equation system for three points may result in no solution. Therefore, the present invention proposes to introduce geometric conditions such as vector perpendicularity, vector parallelism, and multi-point coplanarity for various types of shape targets, establish additional equation systems, and perform position calculation. By reasonably selecting the equation systems, the situations of multiple solutions and incorrect solutions can be avoided.

[0090] The specific implementation methods for establishing constraint equation systems for parallelograms, circles, and triangles in this step are as follows:

[0091] A. Use the distance constraint and parallel constraint of the parallelogram to establish the geometric constraint equation system of the parallelogram image. The specific method is as follows: As shown in Figure 3 (a), take the four corner points of the parallelogram as feature points to describe the target space position; d ij represents the distance between the connecting lines of corner points i and j;

[0092] Establish 6 distance constraint equations:

[0093]

[0094] where i, j = 1, 2, 3, 4 and i ≠ j

[0095] According to the condition that opposite sides of the parallelogram are parallel Establish 4 parallel constraint equations:

[0096]

[0097]

[0098]

[0099]

[0100] For the parallelogram target, a total of 10 equations of distance constraint plus parallel constraint can be established; the four unknowns in the equations are the Z-axis components p wzi of the coordinates of the four corner points of the parallelogram door frame; Select any two equations from the distance constraints plus two equations describing the parallelism of a set of opposite sides to form a well-posed equation system, and calculate the depth values of the pixel coordinates of the four corner points of the parallelogram target;

[0101] B. The projection of a circle is a circle or an ellipse, and the projection of the center of the circle is still the symmetric center of the projected figure, as shown in Figure 3 (b); The connecting lines between the intersection points of the straight line passing through this center and the image edge are all straight lines; Arbitrarily take the intersection points of two straight lines and the image edge to obtain four points p c1 , pc2 , p c3 , p c4 , whose coordinates are P c i = [p cxi , p cyi , 1] T , i = 1, 2, 3, 4, and the coordinates of the four points in the world coordinate system are P i = [X i , Y i , Z i T , and the coordinates in the camera coordinate system are P w i = [P wxi , P wyi , P wzi T , and the coordinates in the pixel coordinate system are P uv i = [u i , v i , 1] T ; Connecting the four points in sequence can form a parallelogram; then it comes back to the problem of constraint modeling for parallelogram-type targets in the present invention. For circular targets, only the diagonal distance is used as the distance constraint, and the following distance constraint equation is established:

[0102]

[0103] i = 1, j = 3 or i = 2, j = 4

[0104] Then, 4 constraint equations are established in the same way as the parallel condition of the opposite sides of the parallelogram in A;

[0105] For circular targets, there are a total of 6 equations for distance constraint and parallel constraint. Select two equations of distance constraint plus any set of two equations describing the parallelism of opposite sides to form a well-posed system of equations, and calculate the depth values of the pixel coordinates of the four characteristic points of the circular target;

[0106] C. For triangular targets, it is known that the side length of the triangle is d ij ; As shown in Figure 3 (c), first, three equations are established according to the distance constraint:

[0107]

[0108] i, j = 1, 2, 3 and i ≠ j

[0109] If the triangle is an equilateral triangle, according to the properties of an equilateral triangle, the line connecting any vertex to the midpoint of the opposite side is perpendicular to the opposite side; the mathematical expression of the vector perpendicular relationship in the Cartesian coordinate system is:

[0110] ​​

[0111] For an equilateral triangle target, there are a total of 4 equations for the distance constraint plus the perpendicular constraint. Select any two equations from the distance constraints and add the equation describing the perpendicular relationship of the vectors to form a well-posed system of equations, and solve for the depth values of the pixel coordinates of the three feature points of the equilateral triangle target;

[0112] If the triangle is a non-equilateral triangle, then solve for the depth values of the three feature points through the three equations established by the distance constraint, and determine the three-dimensional spatial position of the triangle target.

[0113] Step 4: Solve the constraint equations through an improved Newton iteration method to obtain the depth values of the feature points; the specific implementation method is as follows: the input of the improved Newton iteration method is the system of multivariate nonlinear equations constructed in Step 3, and the output is the depth values of the feature points; the improved Newton iteration method with fifth-order convergence is adopted, and the iteration equation is:

[0114]

[0115]

[0116]

[0117] X n is an unknown vector composed of the depth values p of the feature points wzi Y n , Z n are intermediate quantities in the iteration process; considering that the distance between the actual monocular camera and the target is approximately in the range of 1 - 10 meters, set the initial values of all depth values in X n to 200.

[0118] Set the termination condition of the iterative algorithm as:

[0119] ||X n+1 - X n ||1 < 0.01 or η ≥ 20

[0120] where η represents the number of iterations, and the maximum number of iterations is 20; substitute the calculated depth values p of the feature points wzi into the camera projection equation to determine the coordinate point P of each feature point in the camera coordinate system w i:

[0121]

[0122] Step 5: Fuse the camera pose [R|t] obtained by monocular SLAM to obtain the coordinates P of the feature points in the world coordinate system i = [X i , Y i , Z i ​T , where \(i = 1, 2, \ldots, n\), to achieve fast positioning of multi-shaped objects, the mathematical description is as follows:

[0123] P w = RP i + t

[0124] R and t are the rotation matrix and translation vector respectively describing the change in camera pose.

[0125] Those of ordinary skill in the art will realize that the embodiments described herein are for helping readers understand the principles of the present invention and should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Those of ordinary skill in the art can make various other specific deformations and combinations without departing from the essence of the present invention according to these technical revelations disclosed in the present invention, and these deformations and combinations are still within the protection scope of the present invention.

Claims

1. A fast positioning method for multi-shape targets based on target geometric constraints, characterized in that It includes the following steps: Step 1: Use the vision sensor carried by the mobile robot platform to collect images, arrange them in chronological order to form an image sequence S, and obtain the shape type Type of the target in the image through the target detection algorithm; Step 2: According to the shape type Type of the target, select the corresponding image segmentation and feature point extraction method to obtain the coordinates of the target feature points in the pixel coordinate system; the shape type Type of the target includes parallelogram, circle or triangle; Step 3: According to the camera projection equation and the coordinates of the target feature points in the pixel coordinate system, establish a constraint equation set in combination with the target geometric constraint information; the specific implementation method is: A. Establish a geometric constraint equation system for a parallelogram image using distance constraints and parallel constraints of parallelograms. The specific method is as follows: Take the four corner points of the parallelogram as feature points to describe the target spatial position. The coordinates of the four points in the world coordinate system are P i =[X i , Y i , Z i T , i = 1, 2, 3, 4; The coordinates in the camera coordinate system are P w i = [P wxi , P wyi , P wzi T ; At the same time, in the normalized plane, the normalized projection point coordinates of the target are expressed as P c i = [p cxi , p cyi , 1] T ; The coordinates of the target in the pixel coordinate system are P uv i = [u i , v i , 1] T ;​​ d ij represents the distance of the line connecting corner points i and j; Establish 6 distance constraint equations: i, j = 1, 2, 3, 4 and i ≠ j According to the condition that the opposite sides of a parallelogram are parallel Four parallel constraint equations are established: For parallelogram targets, establish a total of 10 equations including distance constraints and parallel constraints; The four unknowns in the equation are the components p of the coordinates of the four corner points of the parallelogram door frame on the Z-axis wzi ; Select any two equations from the distance constraints and add the two equations describing the parallelism of a set of opposite sides to form a well-posed system of equations, and solve for the depth values of the pixel coordinates of the four corner points of the parallelogram target; B. The projection of a circle is a circle or an ellipse, and the projection of the center of the circle is still the center of symmetry of the projected figure; the line connecting the intersection points of the line passing through this center and the image edge is the diameter; take the intersection points of two lines passing through the center of the circle and the image edge to obtain four points p c1 , p c2 , p c3 , p c4 , and its coordinates are P c i = [p cxi , p cyi , 1] T , i = 1, 2, 3, 4, and the coordinates of the four points in the world coordinate system are P i = [X i , Y i , Z i T , and the coordinates in the camera coordinate system are P w i = [P wxi , P wyi , P wzi T , and the coordinates in the pixel coordinate system are P uv i = [u i , v i , 1] T ; the four points connected in sequence can form a parallelogram; for a circular target, only the diagonal distance is used as the distance constraint, and the following distance constraint equation is established:​​ i = 1, j = 3 or i = 2, j = 4 Then establish 4 constraint equations in the same way as the parallel condition of the opposite sides of the parallelogram in A; For circular targets, there are a total of 6 equations including distance constraints and parallel constraints. Select two equations of the distance constraints plus any set of two equations describing the parallelism of the opposite sides to form a well-posed equation set, and solve the depth values of the pixel coordinates of the four feature points of the circular target; C. For a triangular target, the side length of the triangle is known as d ij ; First, establish three equations based on the distance constraint: i, j = 1, 2, 3 and i ≠ j If the triangle is an equilateral triangle, according to the properties of an equilateral triangle, the connection line from any vertex to the midpoint of the opposite side is perpendicular to the opposite side; the mathematical expression of the vector perpendicular relationship in the Cartesian coordinate system is: For equilateral triangle targets, there are a total of 4 equations including distance constraints and perpendicular constraints. Select any two equations of the distance constraints plus the equations describing the vector perpendicular relationship to form a well-posed equation set, and solve the depth values of the pixel coordinates of the three feature points of the equilateral triangle target; If the triangle is a non-equilateral triangle, then solve the depth values of the three feature points through the three equations established by the distance constraints to determine the three-dimensional spatial position of the triangle target; Step 4: Solve the constraint equation set through the improved Newton iteration method to obtain the depth values of the feature points; Step 5: Fuse the camera pose [R|t] obtained by monocular SLAM to obtain the coordinates of the feature points in the world coordinate system, and realize the rapid positioning of multi-shape targets. R and t are the rotation matrix and translation vector describing the change of the camera pose respectively.

2. The fast positioning method for multi-shaped objects based on target geometric constraints according to claim 1, characterized in that, The specific implementation method of the said Step 2 is: Step 21: Perform threshold processing on the detected image to obtain a binary image; Step 22: Obtain the edge information of the target in the image through the adaptive Canny edge extraction algorithm; Step 23: Use the Hough line transform to extract the line features in the parallelogram or triangle edge image, and use the Hough circle transform to extract the arc features in the circular edge image; Step 24: For line features, extract feature corner points through Harris corner detection; for arc features, randomly select two points on the arc edge, and take the intersection points of the connections between the two points and the center of the circle and the opposite arc as the other two feature corner points respectively. If the connection cannot generate an intersection point with the known arc segment, reselect the points; Step 25: Eliminate abnormal feature corner points according to the target category Type and the known target scale information.

3. The rapid positioning method for multi-shaped targets based on target geometric constraints according to claim 1, characterized in that The specific implementation method of step 4 is as follows: The input of the improved Newton iteration method is the multivariate nonlinear equations constructed in step 3, and the output is the depth value of the feature points; The improved Newton iteration method with fifth-order convergence is adopted, and the iteration equation is: X n is an unknown vector composed of the depth value p of the feature point wzi ; Y n , Z n are intermediate quantities in the iterative process; Set the termination condition of the iterative algorithm as: ||X n+1 -X n ||1 < 0.01 or η ≥ 20 where η represents the number of iterations, and the maximum number of iterations is 20; substitute the calculated depth value p of the feature points wzi into the camera projection equation to determine the coordinate point P of each feature point in the camera coordinate system w i:

Images