Method for tracking a sequence of high-speed video images with a sign of self-adapting scale change

Abstract

The application relates to a high-speed video sequence image adaptive scale change mark continuous tracking method, which comprises the following steps: step 1, a fast shape feature extraction algorithm is adopted to process the obtained image sequence, and the pixel coordinates of the artificial mark center are obtained; step 2, an adaptive scale change mark sequence tracking strategy is adopted, that is, according to the radius size of the last frame mark point obtained in real time, the size of the tracking search window of the next frame is dynamically changed, and the backward transmission of the region scale information between adjacent frames of images is realized. Compared with the prior art, the application has the advantages of good accuracy and reliability.

Description

Technical Field

[0001] This invention relates to high-speed video measurement technology, and in particular to a method for continuous tracking of markers that adapts to scale changes in high-speed video sequence images. Background Technology

[0002] High-speed video measurement technology, a non-contact technique that determines the shape and state of a target by acquiring and processing images, has the advantages of not damaging the target and not interfering with the natural state of the object being measured. It can instantly acquire the physical and geometric information of the object in image form. Its image data can be reused and stored long-term, and through analytical photogrammetry, the three-dimensional spatial coordinates of the target points can be obtained, providing various product data based on three-dimensional spatial coordinates. Therefore, this technology is widely used in civil engineering, aerospace, industrial manufacturing, and many other fields.

[0003] To improve the processing speed and measurement accuracy of high-speed video measurement, artificial markers are widely used in the measurement process. The most common measurement scheme involves attaching artificial markers (usually composed of a white solid circle with a black border) to key nodes of the object under test and accurately extracting the center pixel coordinates using a relevant ellipse center fitting algorithm to represent the actual motion state of the object. Based on the initial pixel coordinates of the marker points in the initial frame, tracking methods are used to track the marker points in the sequence of images to obtain the motion state information of the object at each moment. Commonly used tracking methods include phase correlation matching (PC), normalized cross-correlation coefficient matching (NCC), and least squares matching (LSM). LSM tracking is most frequently chosen due to its high tracking accuracy and fast tracking efficiency; for example, the internationally renowned visual measurement software Photomodeler uses the least squares matching algorithm as its automatic marker tracking method.

[0004] However, when using the above tracking method, the size of the search window for matching and tracking markers between adjacent video frames must be specified, and this search window size cannot be dynamically changed during the tracking process. When the object under test moves a long distance in the depth direction within the effective field of view of high-speed video measurement, the size of the image pixel area occupied by the artificial markers will also change significantly. A fixed search window size will cause the above tracking method to fail. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the existing technology and provide a method for continuous tracking of markers that adapts to scale changes in high-speed video sequence images.

[0006] The objective of this invention can be achieved through the following technical solutions:

[0007] According to one aspect of the present invention, a method for adaptive scaling of flags in high-speed video sequence images is provided, the method comprising the following steps:

[0008] Step 1: The acquired image sequence is processed using a fast shape feature extraction algorithm to obtain the pixel coordinates of the artificial marker center;

[0009] Step 2: An adaptive scale-change marker sequence tracking strategy is adopted, which dynamically changes the size of the tracking search window in the next frame based on the real-time acquired marker point radius of the previous frame, thereby realizing the backward transfer of regional scale information between adjacent frames.

[0010] As a preferred technical solution, step 1 specifically includes:

[0011] Step 101: Perform corresponding preprocessing on the image block image of the search window;

[0012] Step 102: Perform edge detection on the preprocessed image block using a topology algorithm based on boundary tracking, and save the set of pixel coordinates of all edge contours.

[0013] Step 103: Perform high-precision fitting on the ellipse in the two-dimensional pixel plane to obtain the pixel coordinates of the artificial mark center.

[0014] As a preferred technical solution, step 101, which involves preprocessing the image block image in the search window, specifically includes:

[0015] Step 1011) Use Gaussian filtering to remove unwanted noise from the image;

[0016] Step 1012) Binarization is used to enhance the contrast of the image block, highlight the white elliptical part of the marker point, and improve the accuracy of edge detection.

[0017] As a preferred technical solution, an index lookup table is established from largest to smallest based on the area size of all edge detection contours. When fitting the ellipse center coordinates based on the contours in the subsequent process, the lookup table is traversed, and the large-area contours are calculated first and then the subsequent condition constraints are judged, which can more quickly determine the required target value.

[0018] As a preferred technical solution, step 103, which involves performing high-precision fitting on the ellipse within the two-dimensional pixel plane, specifically comprises:

[0019] Step 1031) In a two-dimensional plane coordinate system, any ellipse can be represented in algebraic form using the following conic section equation:

[0020] F(a, x) = a·x = ax 2 +bxy+cy2 +dx+ey+f=0 (1)

[0021] In the formula, a = [abcdef] T , x = [x 2 xy y 2 xy 1] T (x, y) represents any point on the ellipse;

[0022] Step 1032) Perform high-precision fitting of the elliptical center coordinates of the pixel plane based on the least squares method, let C contour =[(x1,y1),(x2,y2),…,(x n y n Let )] be the set of pixel coordinates of the edge of the elliptical contour of the obtained image block. Then, according to formula (1), the following optimal estimation objective function can be established:

[0023]

[0024] According to the principle of extrema, to make ε 2 For the value to reach its minimum, the following should be true:

[0025]

[0026] This yields a system of linear equations. Combining the elliptic constraint a + c = 1 and applying the Gaussian elimination method with full principal components, the coefficients a = [abcdef] are obtained. T The value;

[0027] Furthermore, the coordinates of the ellipse's center (x0, y0), the semi-major and semi-minor axes (a0, b0), and the rotation angle θ (the angle between the major axis and the x-axis) of the major axis can be precisely solved using the following formulas:

[0028]

[0029] Where a, b, c, d, e, and f are the coefficients of the ellipse equation in formula (1).

[0030] As a preferred technical solution, the least squares ellipse center pixel coordinates of all detected contours in the image block of the search window are fitted according to the previously established lookup table.

[0031] As a preferred technical solution, the fitting result is determined to be within a 5×5 neighborhood of the center of the search window.

[0032] As a preferred technical solution, step 2 specifically includes:

[0033] 201) Taking the center coordinates of the artificial marker in the i-th frame as the center, expand outwards in all directions by N times the major semi-axis of the ellipse to determine the tracking search window of the (i+1)-th frame.

[0034] 202) Use a fast shape feature extraction algorithm within the tracking search window of the (i+1)th frame to solve for the center coordinates of the target point in the (i+1)th frame;

[0035] 203) The coordinates of the center of the circle and the semi-major axis of the ellipse obtained from solving the i+1 frame are updated and saved, and used as the basis for determining the tracking search window when tracking the target point in the next frame.

[0036] As a preferred technical solution, N is 1.5.

[0037] As a preferred technical solution, the method makes full use of the unique geometric properties of the ellipse, extracts shape features by backward transmission of regional scale information between adjacent frames, and continuously updates the semi-major axis of the ellipse used to determine the tracking search window when tracking the target point in the next frame.

[0038] Compared with existing technologies, this invention can quickly and accurately extract the pixel coordinates of the center of artificial markers through marker shape feature extraction. Furthermore, it can dynamically change the size of the tracking search window in the next frame based on the real-time acquired radius of the marker point in the previous frame, achieving backward transfer of regional scale information between adjacent frames and overcoming the limitations of traditional fixed template matching methods. Through comparative analysis of actual experiments, the tracking method proposed in this invention achieves a single-frame tracking time of approximately 30ms (test image size 1280×1024) and sub-pixel accuracy, outperforming the fixed template matching algorithm used in internationally renowned visual measurement software, thus verifying the efficiency and robustness of the algorithm. Attached Figure Description

[0039] Figure 1 Here is a flowchart of the fast ellipse fitting algorithm;

[0040] Figure 2 Flowchart for adaptive scaling marker sequence tracking;

[0041] Figure 3 A schematic diagram of an experimental scenario for measuring key nodes of a shaking table building structure.

[0042] Figure 4 This is a diagram comparing the results of pixel coordinate tracking.

[0043] Figure 5 This is a diagram illustrating the comparison of deviations in the tracking method results;

[0044] Figure 6 A schematic diagram of a high-speed video measurement experiment scenario for ground testing of long-distance moving objects;

[0045] Figure 7 This is a diagram showing the comparison between the tracking results and the true values ​​of the two tracking methods;

[0046] Figure 8 This is a schematic diagram showing the coordinate deviation of the results from the two tracking methods. Detailed Implementation

[0047] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0048] 1. Introduction

[0049] With the widespread application and development of high-speed video measurement, artificial markers are required to ensure the accuracy of measurement results. Accurate marker sequence tracking has become a research hotspot in high-speed video measurement. Traditional marker tracking methods (such as template matching tracking methods based on least squares matching) require a given search window size for marker matching between adjacent image frames, and this search window size must remain constant during tracking. When the depth direction of the object under test changes significantly within the effective field of view of high-speed video measurement, the size of the image pixel area occupied by the artificial marker will also change significantly. A fixed search window size will cause traditional tracking methods to fail. To solve this problem, this invention proposes a continuous artificial marker tracking method for high-speed video sequence images with adaptive scale changes. Through actual experimental analysis, the proposed tracking method achieves a single-frame tracking time of approximately 30ms (image size 1280×1024) and a tracking accuracy reaching sub-pixel level, outperforming the fixed template matching algorithm used by internationally renowned visual measurement software, thus verifying the efficiency and robustness of this algorithm.

[0050] 2. Artificial Marker Adaptive Scale Change Tracking Method

[0051] 2.1 Fast Shape Feature Extraction Algorithm

[0052] Since the search window image block range during the tracking process is not too large, and the design and production of the artificial markers ensure that the foreground and background are clearly distinguishable, the method of combining mathematical morphology and shape geometric feature extraction can quickly and accurately achieve the recognition and positioning of elliptical markers. Figure 1 This is the overall flowchart of the fast shape feature extraction algorithm proposed in this invention.

[0053] High-speed industrial cameras used in high-speed video measurement may acquire poor image quality due to their short exposure time. Therefore, it is necessary to perform corresponding preprocessing on the image block images in the search window to enhance the contrast of the image blocks and improve the accuracy of edge detection, which is beneficial for subsequent elliptical mark recognition and center coordinate fitting.

[0054] Preprocessing mainly includes two steps: 1) Gaussian filtering: used to remove unnecessary noise in the image; 2) Binarization: enhances the contrast of image blocks, highlights the white elliptical parts of the markers, and improves the accuracy of edge detection.

[0055] Next, edge detection is performed on the preprocessed image blocks using a topology algorithm based on boundary tracking. The pixel coordinates of all edge contours are saved for subsequent use in combining ellipse fitting algorithms with relevant discriminative constraints to obtain the center coordinates of the target points. Furthermore, given the design pattern of artificial markers, the area enclosed by the marker contours within the image block is relatively large. Therefore, based on mathematical morphology, an index lookup table is established from largest to smallest based on the area of ​​all edge detection contours. During subsequent ellipse center coordinate fitting based on the contours, the lookup table is traversed, prioritizing the calculation of large-area contours and subsequent conditional constraint discrimination. This allows for faster iterative discrimination to the desired target value, significantly reducing the algorithm's loop process and thus improving the efficiency of the tracking method.

[0056] Based on the above preprocessing steps, high-precision fitting of the ellipse within the two-dimensional pixel plane is required. In a two-dimensional coordinate system, any ellipse can be represented algebraically using the following conic section equation:

[0057] F(a, x) = a·x = ax 2 +bxy+cy 2 +dx+ey+f=0 (1)

[0058] In the formula, a = [abcdef] T , x = [x 2 xy y 2 xy 1] T (x, y) represents any point on the ellipse. When the random error follows a normal distribution, the least squares method, derived from the maximum likelihood method, is an optimal estimation algorithm that minimizes the sum of squared measurement errors. It is considered one of the most reliable methods for solving unknown quantities from known measurements. Therefore, this invention will use the least squares method to perform high-precision fitting of the ellipse center coordinates of the pixel plane. Let C... contour =[(x1,y1),(x2,y2),…,(x n y nLet )] be the set of pixel coordinates of the edge of the elliptical contour of the obtained image block. Then, according to Formula 1, the following optimal estimation objective function can be established:

[0059]

[0060] According to the principle of extrema, to make ε 2 For the value to reach its minimum, the following should be true:

[0061]

[0062] This yields a system of linear equations. Combining this with the elliptic constraint a + c = 1 and applying methods such as Gaussian elimination with full principal components, the coefficients a = [abcdef] can be obtained. T The values ​​of the ellipse center coordinates (x0, y0), semi-major and semi-minor axes (a0, b0), and the rotation angle θ (the angle between the major axis and the x-axis) can then be precisely solved using the following formulas:

[0063]

[0064] Based on the previously established lookup table, the least squares ellipse center pixel coordinates of all detected contours in the image block of the search window are fitted. Since the center of the fitted ellipse is usually located near the center of the entire search window in the tracking scenario, the fitting result is determined to be within a 5×5 neighborhood of the center of the search window. This allows for further pruning of the algorithm loop process and improvement of the overall tracking efficiency.

[0065] 2.2 Adaptive Scale Change Flag Sequence Tracking Strategy

[0066] Based on the aforementioned fast shape feature extraction algorithm, this invention proposes the following adaptive scale-changing marker sequence tracking strategy: 1) Using the center coordinates of the artificial marker in the i-th frame as the center, expand outwards by 1.5 times the semi-major axis of the ellipse to determine the tracking search window for the (i+1)-th frame; 2) Within the tracking search window of the (i+1)-th frame, use the fast shape feature extraction algorithm to solve for the center coordinates of the target point in the (i+1)-th frame; 3) Update and save the center coordinates and semi-major axis of the ellipse obtained in the (i+1)-th frame, using them as the basis for determining the tracking search window when tracking the target point in the next frame. The specific tracking strategy flowchart is as follows: Figure 2 As shown.

[0067] As can be seen from the algorithm flow proposed in this invention, this algorithm fully utilizes the unique geometric properties of ellipses. Through shape feature extraction via backward propagation of regional scale information between adjacent image frames, it continuously updates the semi-major axis of the ellipse used to determine the tracking search window when tracking the target point in the next frame, thus solving the defect of traditional tracking methods that cannot dynamically change the tracking search window. Simultaneously, since the tracking process in each frame can obviously be viewed as a separate fast ellipse fitting calculation of the search window image block, the tracking in each frame is independent, avoiding the problem of accumulated errors that easily occur in traditional tracking methods. Given that the fitting error of the fast shape feature extraction proposed in this invention is at the sub-pixel level, the error of the tracking method of this invention can also reach the sub-pixel level.

[0068] 3. Experimental verification

[0069] To verify the robustness and reliability of the proposed algorithm, experiments were conducted in both a traditional experimental scenario and an experimental scenario where the size of the marker points continuously changes. In the traditional experimental scenario, the tracking results of the proposed method were first compared with those of a fixed-template LSM tracking method (using the internationally renowned visual measurement software Photomodeler), using manually extracted marker point coordinates as the ground truth. This verified the applicability and reliability of the proposed algorithm under conditions of small scale changes (experimental scenario 1). Secondly, in a long-distance movement experimental scenario where the size of the artificial markers continuously changes (experimental scenario 2), the limitations of the traditional LSM tracking method were demonstrated, and the proposed tracking method was tested, verifying its effectiveness in adaptively tracking artificial markers with varying scales.

[0070] 3.1 Experimental Scenario 1 Experimental Results

[0071] High-precision three-dimensional measurement of the motion state of key structural nodes in civil shaking table experiments using high-speed video measurement is a classic experimental scenario where high-speed video technology is applied. Figure 3 As shown. In this application scenario, the markers attached to key nodes of the building structure have minimal displacement in the depth direction of the effective field of view of the high-speed stereo camera, therefore the size of the markers does not change significantly in the image. This invention compares the proposed adaptive scale change tracking method with the LSM tracking method to verify the applicability and reliability of the algorithm proposed in this invention.

[0072] The image sequence in this verification experiment was 5340 frames long, and each image was 1280×1024 pixels in size. The three methods described above were used to... Figure 3 The center pixel coordinates of the marker points within the red box are tracked and recorded. The tracking results are compared, and the deviations between the two tracking methods and the true values ​​are shown below. Figure 4 and Figure 5 As shown.

[0073] Comprehensive analysis Figure 4 and Figure 5 As can be seen, the tracking method proposed in this invention exhibits good tracking performance in classic scenarios, with overall trends consistent with the LSM tracking method, and both are highly consistent with the manually selected ground truth. Furthermore, the comparison curves of the deviations between the results and the ground truth for the two tracking methods show that the proposed method achieves smaller tracking deviations compared to the LSM tracking method under conditions of significant motion changes, with the overall deviation remaining at the sub-pixel level. Although the tracking method of this invention experiences slight fluctuations during the initial and final static phases, the fluctuation amplitude remains within 0.05 pixels, having minimal impact on subsequent forward intersection calculations. Therefore, the applicability and reliability of the tracking method of this invention in classic experimental scenarios can be verified.

[0074] In addition, the time consumption of the two tracking methods was calculated separately, and the results are shown in Table 1. The total time for tracking 5340 frames of images with a size of 1280×1024 pixels using the tracking method proposed in this invention was 160.25s, with an average tracking time of about 30ms per frame. It is slightly faster than the LSM tracking method in terms of tracking calculation efficiency.

[0075] Table 1

[0076] Total time elapsed (s) 171.43 160.25

[0077] 3.2 Experimental Scenario 2 Experimental Results

[0078] In high-speed video measurement for long-distance moving object ground testing, the spatial three-dimensional coordinates of four circular marker points attached to the model surface are measured with high precision to determine the positions of the model and the artificial markers, such as... Figure 6 As shown.

[0079] During the long-distance movement of the model from far to near, the four artificial markers attached to the model surface will move significantly in the depth direction within the effective field of view of the binocular high-speed camera, resulting in substantial changes in the size of the markers in the image. The fixed-template LSM tracking method commonly used in internationally renowned visual measurement software such as Photomodeler cannot perform full-process tracking in this situation because when the size of the marker to be tracked exceeds the tracking search window size determined by the starting frame, the tracking process will malfunction and be interrupted, or the tracking results will deviate significantly. One feasible solution is to segment the overall tracking process at appropriate marker size points, determine the search window size for each segment, and then perform segmented tracking based on this. However, the selection of appropriate segments and the determination of the tracking search window operations consume a significant amount of processing time and manual effort, and may still not guarantee reliable tracking results. In contrast, the tracking method proposed in this invention can efficiently track the pixel coordinates of the marker points continuously throughout this experimental scenario, while ensuring accurate and reliable tracking results.

[0080] Similarly, this invention uses manually extracted marker pixel coordinates frame by frame as ground truth to compare and verify the tracking performance of the LSM tracking method and the proposed tracking method in this scenario. It is worth noting that this invention divides the entire tracking process into six segments when using the LSM algorithm, and finally concatenates the results to obtain the overall tracking result. The specific comparison of the tracking results and coordinate deviations of the two tracking methods are as follows: Figure 7 and Figure 8 As shown.

[0081] Table 2

[0082] LSM tracking method 0.149856398 0.210975421 Tracking method of the present invention 0.036418028 0.034576298

[0083] Depend on Figure 7It can be seen that the tracking method proposed in this invention maintains a high degree of consistency with the manually selected ground truth value in the experimental scenario where the size of the marker scale changes, demonstrating good tracking performance. In contrast, the LSM tracking method, after segmented tracking processing, although it can also obtain an overall trend that is relatively consistent with the manually selected ground truth value, exhibits a significant tracking deviation at the end of the tracking process. Further comprehensive comparison of the coordinate deviation curves of the two tracking methods reveals that although the tracking method of this invention experiences a slight increase in deviation during marker movement, its coordinate deviation in all directions remains within approximately ±0.1 pixels. In contrast, the LSM tracking method shows a deviation of approximately ±0.5 pixels starting from the middle of the tracking process, and the deviation gradually increases to a maximum of approximately ±1.5 pixels at the end of the tracking process. This is mainly because the pixel size change of the marker in the model accelerates at the end of the tracking process, leading to instability during least-squares matching between adjacent frames. As shown in Table 2, the accuracy evaluation of the adaptive tracking method and the fixed template LSM algorithm of this invention shows that the tracking method of this invention has a high consistency with the manual tracking results. Compared with the fixed template LSM tracking method used by the international visual measurement software Photomodeler, the accuracy of the marker sequence tracking is improved by about 80%. Therefore, the above comparative analysis can further verify the accuracy and stability of the tracking method proposed in this invention.

[0084] The tracking times for the two methods in this experimental scenario were calculated separately. Since the LSM tracking method requires additional time for selecting a suitable segmentation strategy and determining the tracking search window, which typically takes about 2-3 minutes, this time should also be considered in the tracking time calculation for the LSM method. Finally, the total time spent tracking 1048 frames of an image with a size of 1280×1024 pixels is shown in Table 3. Clearly, the tracking method of this invention significantly reduces the overall tracking processing time and greatly improves tracking efficiency.

[0085] Table 3

[0086] Total time elapsed (s) 32.69+120～180 31.47+0

[0087] 4. Conclusion

[0088] This invention addresses the problem that when the depth direction of the object under test changes significantly within the effective field of view of high-speed video measurement, the scale of the image pixel area occupied by the artificial marker changes accordingly, causing traditional tracking methods to fail. It proposes a marker continuous tracking method that adapts to scale changes in high-speed video sequence images. Through experimental analysis in different high-speed video measurement scenarios, and comparison with traditional tracking methods using manually labeled data as the ground truth, the invention effectively verifies that the proposed tracking method has good accuracy and reliability. This method has significant meaning and value in solving engineering application problems related to high-speed video measurement of long-distance moving objects in the depth direction.

[0089] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for adaptive scale-changing marker tracking of high-speed video sequence images, characterized in that, The method includes the following steps: Step 1: The acquired image sequence is processed using a fast shape feature extraction algorithm to obtain the pixel coordinates of the artificial marker center; Step 2: Adopt an adaptive scale change marker sequence tracking strategy, that is, dynamically change the size of the tracking search window in the next frame based on the real-time acquired marker point radius of the previous frame, so as to realize the backward transfer of regional scale information between adjacent frame images. Step 1 specifically includes: Step 101: Perform corresponding preprocessing on the image block image of the search window; Step 102: Perform edge detection on the preprocessed image block using a topology algorithm based on boundary tracking, and save the set of pixel coordinates of all edge contours. Step 103: Perform high-precision fitting on the ellipse in the two-dimensional pixel plane to obtain the pixel coordinates of the artificial mark center; Based on the area size of all edge detection contours, an index lookup table is established from largest to smallest. When fitting the coordinates of the ellipse center based on the contours in the subsequent process, the lookup table is traversed, and the large-area contours are calculated first and the subsequent condition constraints are judged, which can more quickly determine the required target value. Step 2 specifically includes: 201) Taking the center coordinates of the artificial marker in the i-th frame as the center, expand outwards by N times the major semi-axis of the ellipse to determine the tracking search window of the (i+1)-th frame; 202) Use a fast shape feature extraction algorithm within the tracking search window of the (i+1)th frame to solve for the center coordinates of the target point in the (i+1)th frame; 203) Update and save the center coordinates and semi-major axis of the ellipse obtained from the solution of the (i+1)th frame, and use them as the basis for determining the tracking search window when tracking the target point in the next frame; Step 103, which involves performing high-precision fitting of the ellipse within the two-dimensional pixel plane, specifically comprises: Step 1031) In a two-dimensional plane coordinate system, any ellipse can be represented in algebraic form using the following conic section equation: （1） In the formula , , This represents any point on the ellipse; Step 1032) Perform high-precision fitting of the elliptical center coordinates of the pixel plane based on the least squares method, assuming... Given the set of pixel coordinates of the elliptical contour edge of the obtained image block, the following optimal estimation objective function can be established according to formula (1): （2） According to the principle of extrema, to make For the value to reach its minimum, the following should be true: （3） This yields a system of linear equations, which, combined with the elliptic constraint conditions... The linear equation system was solved using the Gaussian elimination method with full principal components, and the equation coefficients were obtained. The value; Therefore, the coordinates of the center of the ellipse Major and minor half-shafts and the angle of rotation of the major axis The exact solution is obtained from the following equations: （4） in The coefficients of the ellipse equation in formula (1); Based on the previously established lookup table, the least squares ellipse center pixel coordinates of all detected contours in the image block of the search window are fitted; the fitting result is used to determine whether it is within the 5×5 neighborhood of the center of the search window.

2. The method for continuous tracking of flags with adaptive scale changes in high-speed video sequence images according to claim 1, characterized in that, Step 101, which involves preprocessing the image block in the search window, specifically includes: Step 1011) Use Gaussian filtering to remove unwanted noise from the image; Step 1012) Binarization is used to enhance the contrast of the image block, highlight the white elliptical part of the marker point, and improve the accuracy of edge detection.

3. The method for continuous tracking of flags with adaptive scale changes in high-speed video sequence images according to claim 1, characterized in that, The value of N is 1.

5.

4. The method for continuous tracking of flags with adaptive scale changes in high-speed video sequence images according to claim 1, characterized in that, The method fully utilizes the unique geometric properties of ellipses, extracts shape features by backward propagation of regional scale information between adjacent frames, and continuously updates the semi-major axis of the ellipse used to determine the tracking search window when tracking the target point in the next frame.

Images