Weighted iteration PnP pose resolving method based on mismatching point pair elimination

Abstract

The invention discloses a weighted iteration PnP pose resolving method based on mismatching point pair elimination. The method comprises the following steps: firstly, solving to obtain an initial pose of a camera through a PnP method according to known m groups of 2D-3D point pairs, then calculating a re-projection error of each group of point pairs according to the initial pose, and calculating respective weight and elimination threshold according to corresponding re-projection error values; and finally, updating the resolving weight of the point pair and the elimination threshold value of the mismatching point pair, and continuously iterating until the conditions are met. According to the method, the self-adaptive updating of the point pair weight and the mismatching point pair rejection threshold value can be realized through the re-projection error, the pose resolving precision of the camera is greatly improved, and the resolving speed is improved to a certain extent. The method is suitable for obtaining the camera pose information through the feature point two-dimensional pixel coordinates obtained through image matching and the corresponding world three-dimensional coordinates.

Description

Technical Field

[0001] The present invention belongs to the technical field of computer vision, and particularly relates to a weighted iterative PnP pose calculation method based on the elimination of mismatched point pairs. Background Art

[0002] The UAV positioning method based on scene matching is an important positioning means under the interference of satellite signals. Feature points in the aerial image and the reference image are extracted, and m groups of 2D-3D feature point pairs are obtained through matching. The pose information of the UAV can be calculated using the PnP principle. However, in actual applications, it is inevitable that there are mismatched cases in the feature point pairs, and direct calculation will affect the final result. Weighted iterative PnP will assign weights to each group of point pairs, but the mismatched point pairs will still participate in the calculation process in each iteration, affecting the calculation accuracy. Summary of the Invention

[0003] Object of the Invention: To solve the problems existing in the above-mentioned prior art, the present invention provides a weighted iterative PnP pose calculation method based on the elimination of mismatched point pairs.

[0004] Technical Solution: The present invention discloses a weighted iterative PnP pose calculation method based on the elimination of mismatched point pairs, which is characterized in that it specifically includes the following steps:

[0005] Step 1: Obtain m feature points of the camera internal parameters, obtain the corresponding 2D-3D matching point pairs of the feature points through matching, and obtain the initial pose of the camera;

[0006] Step 2: Calculate the reprojection error e corresponding to each feature point at the nth iteration, and calculate the weight w of each feature point at the nth iteration based on the reprojection error e and the elimination threshold τ at the previous iteration, and eliminate the feature points with zero weight; i and the previous iteration's elimination threshold τ i ; n-1 and calculate the weight w of each feature point at the nth iteration, i and eliminate the feature points with zero weight;

[0007] Step 3: Calculate the pose using the weighted iterative PnP pose calculation method;

[0008] Step 4: Update the elimination threshold τ n ;

[0009] Step 5: Determine whether the iteration stop condition is satisfied. If so, stop the iteration and obtain the final high-precision pose. Otherwise, increase the iteration count and return to Step 2.

[0010] Furthermore, the reprojection error corresponding to each feature point is calculated according to the following formula:

[0011]

[0012] Among them, i represents the i-th feature point; represents the two-dimensional pixel coordinates actually corresponding to the i-th feature point, p i represents the two-dimensional pixel coordinates obtained by performing an affine transformation on the three-dimensional coordinates of the i-th feature point, p i The expression of is:

[0013] p i = K(RP i + t)

[0014] Among them, R is the rotation matrix of the camera, t is the translation vector of the camera, P i is the coordinate of the i-th feature point in the three-dimensional world coordinate system, and K is the camera internal parameter matrix.

[0015] Furthermore, the weight coefficient corresponding to each feature point is calculated according to the following formula:

[0016]

[0017] Among them, represents the average reprojection error.

[0018] Furthermore, the objective function of the weighted iterative PnP pose solution method is:

[0019]

[0020] Among them, N n represents the total number of matching point pairs during the n-th solution.

[0021] Furthermore, the rejection threshold τ during the n-th iteration is calculated according to the following formula n :

[0022] τ n = α n-1 τ0

[0023] Among them, α represents the convergence coefficient, and τ0 is the preset initial rejection threshold.

[0024] Furthermore, the iteration stop condition is that the average reprojection error reaches the preset threshold or the number of remaining feature points reaches the preset minimum threshold.

[0025] Beneficial effects:

[0026] 1. The present invention can eliminate incorrect points in each iteration process by setting the rejection threshold for mismatched points, avoiding the interference of incorrect points on the results in subsequent iteration processes, reducing the number of point pairs in each iteration, reducing the occupation of computing resources, and improving the accuracy and computing speed of the algorithm.

[0027] 2. The present invention can gradually reduce the rejection threshold of feature points by setting the convergence coefficient of the mismatching point rejection threshold, avoiding excessive dependence on the initial pose solved during the calculation process, which may lead to the rejection of a large number of feature points in the first round, and improving the robustness and calculation accuracy of the algorithm. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 It is a flowchart of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0029] The accompanying drawings, which form a part of the present invention, are used to provide a further understanding of the present invention. The illustrative embodiments and descriptions thereof of the present invention are used to explain the present invention and do not constitute an improper limitation to the present invention.

[0030] The present invention provides a weighted iterative PnP pose solution method based on the rejection of mismatching point pairs. The flowchart of this method is as Figure 1 shown, specifically as follows:

[0031] Step 1: Obtain the camera internal parameters and the two-dimensional pixel coordinates and three-dimensional world coordinates of m groups of 2D-3D matching point pairs, and obtain the initial pose of the camera through the PnP method.

[0032] Step 2: Calculate the reprojection error corresponding to each feature point, and calculate the weight corresponding to each pair of points during the solution through the reprojection error.

[0033] Step 3: Set the initial rejection threshold and convergence coefficient of the mismatching point pairs, and realize the adaptive update of the point pair weight and the mismatching point pair rejection threshold. Continuously iterate until the condition is met to obtain the final high-precision pose result.

[0034] In Step 2, according to the camera internal parameter matrix K and m groups of 2D-3D matching point pairs, the initial pose of the camera is obtained through the PnP method. The three-dimensional world coordinate P corresponding to the i-th pair of points is i affinely transformed to obtain the corresponding two-dimensional pixel coordinate p i :

[0035] p i = K(RP i + t)

[0036] where R is the rotation matrix of the camera, t is the translation vector of the camera, P i = [X i , Y i , Z i T , p i = [u i , v i , and T represents the transpose.

[0037] The homogeneous coordinate representation is:

[0038]

[0039] According to the actual observation point corresponding to the three-dimensional point of the feature point on the image The pixel coordinates of the point can be calculated to get the reprojection error e i , the expression is:

[0040]

[0041] When there are mismatched points, the stability and accuracy of the final result will be affected. According to the reprojection error of each point, different weight coefficients are calculated to improve the accuracy of the solution, as shown below:

[0042]

[0043] In the formula, w i is the weight coefficient of the i-th feature point, N n is the number of point pairs during the nth solution.

[0044] In step 3, the average reprojection error is calculated based on the reprojection error of each set of point pairs

[0045]

[0046] The weight coefficient w when solving the design point pair i as follows:

[0047]

[0048] When the weight is 0, it means that the matching point pair corresponding to the feature point is removed, τ n-1 It represents the rejection threshold at the n-1th iteration. The expression of the rejection threshold is as follows:

[0049] τ n =α n-1 τ0

[0050] Among them, α is the convergence coefficient, and τ0 is the preset initial rejection threshold.

[0051] The elimination threshold is updated after each solution, and the iteration is stopped when the average reprojection error reaches the required requirement or the number of remaining feature points has been eliminated to the minimum number required for the solution.

[0052] After experimental verification, for the same data, this method has higher solution accuracy compared with traditional methods such as EPnP (Efficient Perspective-n-Point), UPnP (Uncalibrated Perspective-n-Point), and weighted iterative PnP.

[0053] In addition, it should be noted that the various specific technical features described in the above specific embodiments can be combined in any appropriate manner without conflict. To avoid unnecessary repetition, the present invention does not separately describe various possible combination methods.

Claims

1. A weighted iterative PnP pose solving method based on elimination of mismatched point pairs, characterized in that: The specific steps include: Step 1: Get the m feature points of the camera's intrinsic parameters, obtain the 2D-3D matching point pairs corresponding to the feature points through matching, and get the initial pose of the camera; Step 2: Calculate the reprojection error e corresponding to each feature point at the nth iteration i , based on the reprojection error e i and the removal threshold τ at the previous iteration n-1 , calculate the weight w of each feature point at the nth iteration i , remove feature points with zero weight; Step 3: Calculate the pose using the weighted iterative PnP pose solution method; Step 4: Update the rejection threshold τ n ; Step 5: Determine whether the iteration stop condition is met. If so, stop the iteration to obtain the final high-precision pose. Otherwise, increase the number of iterations by step 2.

2. According to claim 1, a weighted iterative PnP pose solving method based on elimination of mismatched point pairs is characterized in that: The reprojection error corresponding to each feature point is calculated according to the following formula: Among them, i represents the i-th feature point; Indicates the two-dimensional pixel coordinates actually corresponding to the i-th feature point, p i represents the two-dimensional pixel coordinates obtained after affine transformation of the three-dimensional coordinates of the i-th feature point, p i The expression is: p i =K(RP i +t) Among them, R is the rotation matrix of the camera, t is the translation vector of the camera, P i is the coordinate of the i-th feature point in the three-dimensional world coordinate system, and K is the camera intrinsic parameter matrix.

3. The weighted iterative PnP pose solving method based on elimination of mismatched point pairs according to claim 1, characterized in that: The weight coefficient corresponding to each feature point is calculated according to the following formula: in, represents the average reprojection error.

4. The weighted iterative PnP pose solving method based on elimination of mismatched point pairs according to claim 1, characterized in that: The objective function of the weighted iterative PnP pose solution method is: Among them, N n Indicates the total number of matching point pairs during the nth solution.

5. The weighted iterative PnP pose solving method based on elimination of mismatched point pairs according to claim 1, characterized in that: The elimination threshold τ at the nth iteration is calculated according to the following formula n : t n =a n-1 t0 Among them, α represents the convergence coefficient, and τ0 is the preset initial rejection threshold.

6. The weighted iterative PnP pose solving method based on elimination of mismatched point pairs according to claim 1, characterized in that: The iteration stopping condition is that the average reprojection error reaches a preset threshold or the number of remaining feature points reaches a preset minimum threshold.

Images