Image processing method for improving over-the-ground high-speed tracking precision

Abstract

The invention relates to the field of image processing, solves the problem that the tracking precision is reduced due to the influence of the matching quality of images and feature points in the high-speed tracking process, and improves the stable tracking capability to the ground. The method comprises: 1, capturing an image target and obtaining an initial tracking point; 2, taking a limited number of continuous image frames as image processing groups, taking a first frame in each image processing group as a reference frame, and taking the last frame as a matching frame; 3, performing image enhancement preprocessing on the images of the image processing group; 4, performing target edge inner neighborhood sub-pixel Harris corner detection on the image, and generating an FREAK corner feature vector; 5, performing feature point matching on the matching frame and the reference frame, estimating an affine matrix, and transforming the tracking point of the reference frame to generate a new matching frame tracking point; and 6, taking the last frame of image in the above step as a reference frame of the new image processing group, taking a plurality of image frames after the reference frame as matching frames of the new image processing group, and repeating the above steps 3-5 to obtain a tracking point of each frame of image. According to the method, more strong angular points are provided, the accuracy of the affine matrix is ensured, and the tracking drift probability is reduced.

Description

Technical Field

[0001] The present invention relates to the field of image processing, and in particular to an image processing method for improving the accuracy of high-speed ground tracking. Background Art

[0002] With the continuous development of digital image processing technology, high frame rate tracking technology is also constantly developing and progressing. The commonly used high frame rate tracking method is the feature point matching method. This method first obtains the corner points of the target in each frame image, generates feature descriptors based on the corner points, then matches the feature descriptors to estimate the affine matrix, and finally updates the tracking points. The core of this method is feature point matching, which is affected by image quality and corner point distribution, reducing the registration accuracy, and thus affecting the tracking accuracy. Therefore, it is necessary to improve image quality and optimize corner point distribution before corner point detection. For example, Chinese patent 2024101760960 discloses "Target Tracking Method, Device, System and Medium Based on Super-Resolution Image", and the authorization announcement number is CN117726656B. The method includes: obtaining a first image and determining a transformation matrix about the first image; correcting the first image according to the transformation matrix to obtain a second image; performing super-resolution reconstruction on the second image to obtain a first super-resolution image; and tracking the target area in the first super-resolution image. This method has a large amount of data processing and low processing efficiency, and can only be used for tracking stationary or slow-moving targets. Summary of the invention

[0003] The purpose of the present invention is to provide an image processing method for improving the accuracy of high-speed ground tracking, solve the problems of high frame rate image quality and low corner point distribution registration accuracy, and improve the tracking accuracy of the feature point matching tracking method.

[0004] The technical solution adopted by the present invention is:

[0005] The image processing method for improving the accuracy of high-speed ground tracking described in the present invention comprises the following steps:

[0006] 1) The image acquisition device acquires multiple frames of continuous images, uses the capture algorithm to capture the image target, and obtains the initial tracking point;

[0007] 2) The first frame is used as the reference frame, and a limited number of image frames after the reference frame are used as matching frames. The reference frame and the matching frame together constitute an image processing group.

[0008] 3) Performing image enhancement preprocessing on the images of the image processing group;

[0009] c) Preprocess the image by first using median filtering to remove noise.

[0010] d) Applying adaptive histogram equalization (CLAHE) to enhance the brightness and texture of objects in the image;

[0011] 4) Perform Harris corner point detection on the neighborhood of the target edge of the image and generate a FREAK corner point feature vector;

[0012] e) obtaining the target contour and target area of ​​the image;

[0013] f) Divide the image into M×N sub-blocks, mark the sub-blocks intersecting with the target contour as edge blocks, and calculate the maximum corner point response value of each edge block;

[0014] g) Setting a threshold k, obtaining a set of candidate corner points that are greater than k% of the maximum corner point response value, removing corner points outside the target area, and taking the first five points of each corner point set as the final sub-pixel Harris corner points of the edge sub-block;

[0015] h) Generate a FREAK feature vector for each corner point in the inner edge region.

[0016] 5) Match the feature points of each matching frame with the reference frame, estimate the affine matrix, use the affine matrix to transform the tracking points of the reference frame, and generate new tracking points of the matching frame;

[0017] 6) Take the last frame of the image processing group obtained in the above step 5) as the reference frame of the new image processing group, and a limited number of consecutive image frames after the reference frame as the matching frames of the new image processing group, and repeat the above steps 3) to 5) until the tracking points of each frame are obtained.

[0018] Preferably, in step 5), the transformation of the tracking point of the reference frame is to generate new tracking point coordinates of the matching frame by multiplying the original tracking point coordinates of the reference frame by the affine matrix.

[0019] Preferably, in step 2), the number of matching frames in the image processing group is positively correlated with the frame rate at which the image acquisition device acquires images.

[0020] Compared with the prior art, the present invention has the following beneficial effects:

[0021] (1) Improve image quality before corner point detection and provide more strong corner points;

[0022] (2) A method for obtaining corner points is designed to ensure the accuracy of the affine matrix;

[0023] (3) The reference frame is replaced at regular intervals to significantly reduce the probability of tracking drift. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1The present invention is a flow chart of an image processing method for improving the accuracy of high-speed ground tracking. DETAILED DESCRIPTION

[0025] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0026] The present invention provides an image processing method for improving the tracking accuracy of high-speed ground tracking, which is used to improve the tracking accuracy of high-frame-rate ground target tracking. The process of the method is as follows: Figure 1 As shown, the following steps are included:

[0027] 1) The image acquisition device acquires multiple frames of continuous images, uses the capture algorithm to capture the image target, and obtains the initial tracking point;

[0028] 2) The first frame image is used as the reference frame, and a limited number of image frames after the reference frame are used as matching frames. The reference frame and the matching frames together constitute an image processing group. The number of matching frames in the image processing group is positively correlated with the frame rate of the image acquisition device. For example, when the frame rate is 400, the number of matching frames is selected to be 100 or 200; when the frame rate is 200, the number of matching frames is selected to be 50 or 100.

[0029] 3) Performing image enhancement preprocessing on the images of the image processing group;

[0030] e) Preprocess the image by first using median filtering to remove noise.

[0031] f) Apply adaptive histogram equalization (CLAHE) to enhance the brightness and texture of objects in the image;

[0032] 4) Perform Harris corner point detection on the neighborhood of the target edge of the image and generate a FREAK corner point feature vector;

[0033] i) obtaining the target contour and target area of ​​the image;

[0034] j) Divide the image into M×N sub-blocks, mark the sub-blocks intersecting with the target contour as edge blocks, and calculate the maximum corner point response value of each edge block;

[0035] k) setting a threshold k, obtaining a set of candidate corner points that are greater than k% of the maximum corner point response value, removing corner points outside the target area, and taking the first 5 points of each corner point set as the final sub-pixel Harris corner points of the edge sub-block;

[0036] l) Generate a FREAK feature vector for each corner point in the inner edge region.

[0037] 5) Match the feature points of each frame of the matching frame with the reference frame, estimate the affine matrix, and use the affine matrix to transform the tracking points of the reference frame to generate new tracking points of the matching frame; the transformation of the tracking points of the reference frame is to multiply the original tracking point coordinates of the reference frame by the affine matrix to generate new tracking point coordinates of the matching frame.

[0038] 6) Take the last frame of the image processing group obtained in the above step 5) as the reference frame of the new image processing group, and a limited number of consecutive image frames after the reference frame as the matching frames of the new image processing group, and repeat the above steps 3) to 5) until the tracking points of each frame are obtained.

[0039] The present invention is specifically described in detail with reference to the acquired real-time image.

[0040] 1) The tracker acquires high frame rate ground target images (W*H resolution) in real time;

[0041] 2) First, the first frame image is used as the reference frame to determine the tracking point coordinates (x, y), and the reference frame is preprocessed for image enhancement. The sub-pixel Harris corner point detection in the neighborhood of the target edge is performed to generate the FREAK corner point feature vector;

[0042] a) First, use 3*3 template median filtering to remove noise, and then apply adaptive histogram equalization (CLAHE) to improve the target visual quality and enhance the target brightness and texture;

[0043] b) Use the Canny operator to obtain the target contour, and the points inside the contour constitute the target area;

[0044] c) Divide the image into W / 10×N / 10 sub-blocks, mark the sub-blocks intersecting with the contour as edge blocks, calculate the maximum corner point response value of each edge block, and perform non-maximum suppression in the 10*10 area;

[0045] d) Set the threshold to 0.1, obtain the candidate corner point set that is greater than the maximum corner point response value * 0.1, remove the corner points outside the target area, and take the first 5 points of each remaining corner point set as the final sub-pixel Harris corner points of the edge sub-block;

[0046] e) Generate the FREAK feature vector for each corner point in the inner neighborhood of the edge.

[0047] 3) The 99 consecutive frames following the reference frame are used as matching frames. The matching frames are also enhanced first, and then the sub-pixel corner points in the edge neighborhood are detected, and finally the FREAK feature vector is generated;

[0048] 4) Perform feature matching between the matching frame and the reference frame, estimate the affine matrix, use the affine matrix to transform the tracking points of the reference frame, and generate new tracking points of the matching frame;

[0049] 5) Take the matching frame of the last frame as the first frame image, that is, the reference frame, and repeat steps 2-4 until the tracking points of each frame image are obtained.

Claims

1. An image processing method for improving high-speed ground tracking accuracy, characterized in that: The following steps are involved: 1) The image acquisition device acquires multiple frames of continuous images, uses the capture algorithm to capture the image target, and obtains the initial tracking point; 2) The first frame is used as the reference frame, and a limited number of image frames after the reference frame are used as matching frames. The reference frame and the matching frame together constitute an image processing group. 3) Performing image enhancement preprocessing on the images of the image processing group; a) Preprocess the image by first performing median filtering to remove noise. b) Apply adaptive histogram equalization (CLAHE) to enhance the brightness and texture of objects in the image; 4) Perform Harris corner point detection on the neighborhood of the target edge of the image and generate a FREAK corner point feature vector; a) obtaining the target contour and target area of ​​the image; b) Divide the image into M×N sub-blocks, mark the sub-blocks intersecting with the target contour as edge blocks, and calculate the maximum corner point response value of each edge block; c) Setting a threshold k, obtaining a set of candidate corner points that are greater than k% of the maximum corner point response value, removing corner points outside the target area, and taking the first five points of each corner point set as the final sub-pixel Harris corner points of the edge sub-block; d) Generate the FREAK feature vector for each corner point in the edge region. 5) Match the feature points of each matching frame with the reference frame, estimate the affine matrix, use the affine matrix to transform the tracking points of the reference frame, and generate new tracking points of the matching frame; 6) Take the last frame of the image processing group obtained in the above step 5) as the reference frame of the new image processing group, and a limited number of consecutive image frames after the reference frame as the matching frames of the new image processing group, and repeat the above steps 3) to 5) until the tracking points of each frame are obtained.

2. The image processing method for improving high-speed ground tracking accuracy according to claim 1, characterized in that: In step 5), the tracking point transformation of the reference frame is to generate new tracking point coordinates of the matching frame by multiplying the original tracking point coordinates of the reference frame by the affine matrix.

3. The image processing method for improving high-speed ground tracking accuracy according to claim 1 or 2, characterized in that: In step 2), the number of matching frames in the image processing group is positively correlated with the frame rate at which the image acquisition device acquires images.

Images