Multi-template collaborative preferential matching target tracking method

Abstract

The invention discloses a multi-template collaborative preferential matching target tracking method, which comprises the following steps of: based on known target identification information, detecting a plurality of feature points in an area around a target, respectively establishing a feature point template and a target template, and simultaneously establishing a relative position relationship between the plurality of feature points and the target; respectively matching the feature point template established by the current frame image and the template of the target with the original image to be matched and the reduced image to be matched, preferentially selecting the matched feature point according to the matching coefficient, and obtaining the target feature point based on the fact that the relative position relationship between the feature point and the target in the actual scene is kept unchanged. The target position can be indirectly obtained through calculation according to the matched feature points; and then, according to a template matching coefficient of the target, allocating different weights to a target position obtained based on feature point matching calculation and a target position obtained based on template matching of the target, and finally obtaining a tracking position of the target and restoring the tracking position to an original image. According to the invention, the matching precision is improved.

Description

Technical Field

[0001] The invention relates to a multi-template collaborative optimal matching target tracking method, belonging to the technical field of precision guidance. Background Art

[0002] The seeker on the infrared-visible light air-to-ground guided weapon can directly receive the infrared radiation of the target and the visible light information reflected by the target, and has the characteristics of high spatial resolution and high guidance accuracy. However, during the terminal flight of the infrared-visible light air-to-ground guided weapon, the image target scale changes greatly, and the traditional algorithm that uses image template matching to achieve target tracking has poor tracking accuracy, making it difficult to achieve ideal terminal target tracking.

[0003] Commonly used target tracking algorithms based on image template matching include real-time updating of image templates, establishing a template library by scaling templates, and using templates in the template library for template matching. The target tracking algorithm that updates image templates in real time is to establish a template at the matching position in each frame of the image, and then use the template established in the previous frame of the image to match the image to be matched, and cycle in sequence to achieve target tracking. This type of method has poor matching effect when the target is blocked, and due to the real-time update of the template, matching cumulative errors will be generated, causing tracking drift; if the template is updated adaptively, there is a problem with the template update strategy, and when the scale of the target image changes greatly, the template matching accuracy is low; if the template size remains unchanged, this method cannot adapt to the scale change of the target, and if the template size gradually increases with the target, it takes a long time. By scaling the template to establish a template library and use the template in the template library for template matching, it is necessary to first establish a template library, match all templates in the template library with the image to obtain the optimal matching result, which is time-consuming, and the matching effect is poor when the target is blocked. It is not suitable for terminal target tracking of infrared-visible light air-to-ground guided weapons.

[0004] Feature-based template matching and grayscale-based template matching can process a single template and match it to the target position in the most ideal state. However, when the target is occluded to a large extent, rotated at a large angle, or the scale of the target changes greatly, single template matching based on features or grayscale cannot achieve target tracking well. When a mismatch occurs, it will cause the target to deviate from the target. Therefore, single template matching has certain limitations in achieving target tracking. Summary of the invention

[0005] The technical problem solved by the present invention is: to overcome the deficiencies of the prior art, to provide a multi-template collaborative optimal matching target tracking method, and to improve the matching accuracy.

[0006] The technical solution of the present invention is:

[0007] The present invention discloses a multi-template collaborative optimal matching target tracking method, comprising:

[0008] S1. According to the target recognition information, several target feature points are obtained through feature point detection;

[0009] S2, establishing the relative position relationship between a number of target feature points and the target center point;

[0010] S3, establishing several feature point templates and target templates;

[0011] S4, performing NCC template matching on a plurality of feature point templates and a target template with the image to be matched, respectively, to obtain matching coefficients of a plurality of feature points and a matching coefficient of a target point;

[0012] S5. Select multiple matching feature points according to the matching coefficients of the feature point templates;

[0013] S6. Calculate the coordinates of the matching feature target point based on the relative position relationship and multiple matching feature points;

[0014] S7, based on the target's own template, matching to obtain the coordinates of the target point itself;

[0015] S8, according to the matching coefficient of the feature point and the matching coefficient of the target point, respectively assigning different weights to the matching feature target point coordinates and the own target point coordinates to obtain the tracking coordinates;

[0016] S9. Update the target information according to the tracking coordinates, and repeat steps S1 to S8 until the target tracking is completed.

[0017] Furthermore, in the above method, the feature point detection in step S1 is specifically:

[0018] Determine a feature point detection rectangular area, wherein the feature point detection rectangular area is: the coordinates of the upper left corner of the rectangle (m, m), and the coordinates of the lower right corner of the rectangle (Mm, Nm), where M and N are the width and height of the image, and m is the width and height of the feature point template;

[0019] In the feature point detection rectangular area, the pixel points whose absolute difference between the pixel value and the pixel value of the surrounding area is greater than 55 are selected as feature points.

[0020] Furthermore, in the above method, in step S1, feature point detection adopts Surf feature point detection algorithm.

[0021] Furthermore, in the above method, in step S2, the relative position relationship between the plurality of target feature points and the target center point is specifically:

[0022] Connect several target feature points in sequence to form an outer contour;

[0023] Calculate the angle between the line connecting the target center point and each target feature point and the outer contour;

[0024] Furthermore, in the above method, in step S3, the constructed feature point template and the target template itself are of the same size; the width and height of the template are both m, m=1.3-2.5D, and D is the diameter of the target circumscribed circle.

[0025] Furthermore, in the above method, in step S4, the NCC template matching calculation formula is:

[0026]

[0027] Where T(x, y) is the pixel value of the template image, I(x+u, y+v) is the pixel value of the target image at position (u, v), is the average value of the template image, is the average value of the window corresponding to the target image at position (u, v), and NCC(u, v) is the NCC value.

[0028] Furthermore, in the above method, in step S4, the images to be matched are the original image and the reduced image, the reduced image is the original image reduced by a certain multiple with the target point as the center, and the reduction multiple is the ratio of the projectile-target distance corresponding to the template image to the projectile-target distance corresponding to the image to be matched; the pixel value 0 is filled around the reduced image to the size of the original image.

[0029] Furthermore, in the above method, in step S5, multiple matching feature points are selected according to the matching coefficients of the feature point templates, specifically: matching feature points corresponding to the N largest matching coefficients are selected from the multiple matching coefficients; N≥2.

[0030] Furthermore, in the above method, in step S8, different weights are assigned to the matching feature target point coordinates and the own target point coordinates according to the matching coefficients of the feature points and the matching coefficients of the target points, respectively, to obtain the tracking coordinates, and the calculation formula is:

[0031]

[0032] Among them, S pmin is the minimum value among several feature point matching coefficients, S n is the matching coefficient of the target’s own template; (x p ,y p ) is the coordinate of the target point indirectly obtained based on the matching feature points, (x n ,y n ) is the coordinate of the matching target point directly obtained based on the target template itself, and (x, y) is the tracking coordinate.

[0033] Furthermore, in the above method, when the target size is larger than the target template size, the weight of the target point coordinates is 0.

[0034] The beneficial effects of the present invention and the prior art are:

[0035] (1) The present invention can resist partial occlusion of the target. When the target is mostly occluded, the target position obtained based on the target template matching may have a large error. In order to keep the tracking target from being lost, the target position indirectly obtained based on feature point matching can be used for continuous target tracking.

[0036] (2) High target tracking accuracy. The final position of the tracked target is determined by both the target position obtained by matching the target template and the target position obtained indirectly by matching feature points, with high accuracy.

[0037] (3) Adaptable to the change of target scale at the end of bombing. Since the target scale changes greatly at the end of bombing, the image is adaptively reduced according to the distance between the bomb and the target during template matching to reduce the change of target scale.

[0038] (4) It can resist the rotation of the target image perspective. When the image is rotated at a certain angle, the target position indirectly obtained by detecting feature points and matching them can correct the target position obtained by matching the target template itself, thereby improving the matching accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 It is a multi-template collaborative matching target tracking graph of the present invention;

[0040] Figure 2 It is the multi-target feature point and template map of the present invention;

[0041] Figure 3 It is a schematic diagram of the multi-tracking final segment image of the present invention;

[0042] Figure 4 This is the reduced image of the multi-tracking final stage of the present invention. DETAILED DESCRIPTION

[0043] The present invention is further described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0044] The present invention discloses a multi-template collaborative optimal matching target tracking method, comprising:

[0045] S1. According to the target recognition information, several target feature points are obtained through feature point detection;

[0046] S2, establishing the relative position relationship between a number of target feature points and the target center point;

[0047] S3, establishing several feature point templates and target templates;

[0048] S4, performing NCC template matching on a plurality of feature point templates and a target template with the image to be matched, respectively, to obtain matching coefficients of a plurality of feature points and a matching coefficient of a target point;

[0049] S5. Select multiple matching feature points according to the matching coefficients of the feature point templates;

[0050] S6. Calculate the coordinates of the matching feature target point based on the relative position relationship and multiple matching feature points;

[0051] S7, based on the target's own template, matching to obtain the coordinates of the target point itself;

[0052] S8, according to the matching coefficient of the feature point and the matching coefficient of the target point, respectively assigning different weights to the matching feature target point coordinates and the own target point coordinates to obtain the tracking coordinates;

[0053] S9. Update the target information according to the tracking coordinates, and repeat steps S1 to S8 until the target tracking is completed.

[0054] Preferably, the feature point detection in step S1 is specifically:

[0055] Determine a feature point detection rectangular area, wherein the feature point detection rectangular area is: the coordinates of the upper left corner of the rectangle (m, m), and the coordinates of the lower right corner of the rectangle (Mm, Nm), where M and N are the width and height of the image, and m is the width and height of the feature point template;

[0056] In the feature point detection rectangular area, the pixel points whose absolute difference between the pixel value and the pixel value of the surrounding area is greater than 55 are selected as feature points.

[0057] Preferably, in step S1, feature point detection adopts Surf feature point detection algorithm.

[0058] Preferably, in step S2, the relative position relationship between the plurality of target feature points and the target center point is specifically:

[0059] Connect several target feature points in sequence to form an outer contour;

[0060] Calculate the angle between the line connecting the target center point and each target feature point and the outer contour;

[0061] Preferably, in step S3, the constructed feature point template and the target template are of the same size; the width and height of the template are both m, m=1.3-2.5D, and D is the diameter of the target circumscribed circle.

[0062] Preferably, in step S4, the NCC template matching calculation formula is:

[0063]

[0064] Where T(x, y) is the pixel value of the template image, I(x+u, y+v) is the pixel value of the target image at position (u, v), is the average value of the template image, is the average value of the window corresponding to the target image at position (u, v), and NCC(u, v) is the NCC value.

[0065] Preferably, in step S4, the images to be matched are the original image and the reduced image, the reduced image is the original image reduced by a certain multiple with the target point as the center, and the reduction multiple is the ratio of the projectile-target distance corresponding to the template image to the projectile-target distance corresponding to the image to be matched; the pixel value 0 is used to fill the area around the reduced image to the size of the original image.

[0066] Preferably, in step S5, a plurality of matching feature points are selected according to the matching coefficients of a plurality of feature point templates, specifically: matching feature points corresponding to N largest matching coefficients are selected from the plurality of matching coefficients; N≥2.

[0067] Preferably, in step S8, different weights are assigned to the matching feature target point coordinates and the own target point coordinates according to the matching coefficients of the feature points and the matching coefficients of the target points, respectively, to obtain the tracking coordinates, and the calculation formula is:

[0068]

[0069] Among them, S pmin is the minimum value among several feature point matching coefficients, S n is the matching coefficient of the target’s own template; (x p ,y p ) is the coordinate of the target point indirectly obtained based on the matching feature points, (x n ,y n ) is the coordinate of the matching target point directly obtained based on the target template itself, and (x, y) is the tracking coordinate.

[0070] Preferably, when the target size is larger than the target template size, the weight of the target point coordinates is 0.

[0071] Example

[0072] This embodiment adopts multiple positioning such as indirect positioning of feature points and direct positioning of target points, and scales the image to be matched, and obtains the final target point coordinates through NCC template matching, thereby achieving accurate and stable target tracking when the image target scale changes greatly during the final stage of the flight of infrared-visible light air-to-ground guided weapons, solving the problem of target tracking loss due to partial occlusion of the target, and improving the target tracking accuracy through indirect and direct positioning, which can adapt to target tracking with large changes in image target scale, and can resist the rotation of the target image viewing angle, and has high stability and tracking accuracy.

[0073] When infrared-visible light air-to-ground guided weapons are in the final stage of flight, the onboard seeker can receive the infrared radiation of the target and the visible light information reflected by the target. However, in the final stage of flight of infrared-visible light air-to-ground guided weapons, the image target scale changes greatly, and the target is continuously enlarged until the target fills the field of view. At this time, the tracking will drift or jump, resulting in a decrease in target tracking accuracy. Considering that the image scale changes rapidly in the final stage of tracking and the target is gradually enlarged, the present invention proposes dual positioning based on the target and feature points, and adaptively reduces the image by a coefficient, thereby improving the accuracy of target tracking.

[0074] The present embodiment provides a multi-template collaborative optimal matching target tracking method, based on known target recognition information, detects multiple feature points (the number of feature points must be greater than or equal to 2) in the area around the target, establishes the relative position relationship between the feature points and the target, and establishes the feature point template and the target template itself. The feature point template and the target template itself are matched with the original image and the reduced image respectively, and the reduction coefficient is determined by the missile-target distance calculation and adaptively changed according to the matching coefficient. The matching feature points are selected based on the matching coefficient, and the relative position relationship between the feature point and the target remains unchanged in the actual scene, and the target point can be indirectly obtained according to the relative relationship calculation. Then, according to the target template matching coefficient, different weights are assigned to the target position indirectly obtained based on the feature point matching calculation and the target position obtained based on the target template matching, and the final target tracking position is obtained and restored to the original image. The above process is repeated later to achieve stable target tracking. It solves the problem that the commonly used tracking method is not suitable for scale and perspective changes, improves the matching accuracy, and can be used for infrared imaging air-to-ground guided weapon terminal target tracking, with good application prospects.

[0075] like Figure 1 As shown, the present invention provides a multi-template collaborative optimal matching target tracking method, comprising the following steps:

[0076] S1: Based on the known target recognition information, detect the feature points P1, P2, P3 around the target, establish the relative position relationship between each feature point and the target N, and simultaneously establish multiple feature point templates Mask P1, Mask P2 , Mask P3 And the target's own template Mask N ;

[0077] S2: Mask the feature point template P1 , Mask P2 , Mask P3 And the target's own template Mask N Perform NCC template matching with the image to be matched respectively:

[0078] S3: According to the matching coefficients of multiple feature points, the matching feature points are selected optimally, and the matching target points are indirectly obtained by calculation based on the relative position between the feature points and the target being kept unchanged;

[0079] S4: According to the target template matching coefficient and the minimum feature point matching coefficient, the matching target points indirectly obtained based on the matching feature points and the target template Mask obtained based on the target itself are matched. N Different weights are assigned to the directly obtained matching target points;

[0080] S5: The final target point is obtained by weight distribution calculation, and then the coordinates of the target point are restored to the original image.

[0081] In step S1, feature point detection uses the Surf feature point detection algorithm to locate the feature points. Multiple feature point templates Mask are established based on the detected feature points P1, P2, and P3. P1 , Mask P2 , Mask P3 Same size, such as Figure 2 As shown in the figure, the pink box is the feature point and its template, and the red box is the known target point and its template. The relative position relationship between the feature point and the target N is a geometric relationship, and its geometric relationship maintains the original ratio during scale scaling, which can indirectly locate the target point. The template and the image are matched by NCC template, and the NCC template matching calculation formula is:

[0082]

[0083] Where T(x, y) is the pixel value of the template image, I(x+u, y+v) is the pixel value of the target image at position (u, v), is the average value of the template image, is the average value of the window corresponding to the target image at position (u, v), and NCC(u, v) is the NCC value.

[0084] In steps S2-S5, firstly, the target template Mask NPerform NCC template matching with the original image to be matched and the image reduced by r times, and the matching targets are N′ n and N″ n , the corresponding matching coefficient is s′ n and s″ n The reduction coefficient r is determined by the ratio of the projectile-target distance between the current frame image and the image to be matched, and is executed in subsequent frames in a loop.

[0085] When s′ n >s″ n When the target itself is the template Mask N The matching coefficient with the original image to be matched is high, and then the feature point template Mask P1 , Mask P2 , Mask P3 The NCC templates are matched with the original images to be matched, and the matching feature points obtained are P′1, P′2, P′3, and the matching coefficients are s p1 ,s p2 ,s p3 , take s p1 ,s p2 ,s p3 The minimum value is s pmin According to the relative position relationship between feature points P1, P2, P3 and target N, the target point N can be indirectly obtained by calculating the matching feature points P′1, P′2, P′3. P (x P ,y P ), according to the matching coefficient s′ n and pmin The coordinate point N of the matching target point indirectly obtained based on the matching feature point P (x P ,y P ) and Mask based on the target's own template N Directly obtained matching target point N′ n (x n ,y n ) is used to distribute the coefficient weights, and the final target point coordinates (x, y) are as follows:

[0086]

[0087] When s′ appears for the first time n <s″ n When s′ n >s″ n strategy execution;

[0088] If and only if s′ appears twice in a row n <s″ nWhen the target's own template Mook is N The matching coefficient is higher than that of the image to be matched, and then the feature point template Mask P1 , Mask P2 , Mask P3 The NCC template matching is performed with the image to be matched which is reduced by r times, and the matching feature points obtained are P′1, P′2, P′3, and the matching coefficients are s p1 ,s p2 ,s p3 , take s p1 ,s p2 ,s p3 The minimum value is s pmin According to the relative position relationship between feature points P1, P2, P3 and target N, the target point N can be indirectly obtained by calculating the matching feature points P′1, P′2, P′3. P (x P ,y P ), according to the matching coefficient s″ n and pmin The coordinate point N of the matching target point indirectly obtained based on the matching feature point P (x P ,y P ) and Mask based on the target's own template N Directly obtained matching target point N′ n (x n ,y n ) is used to distribute the coefficient weights, and the final target point coordinates (x, y) are as follows:

[0089]

[0090] Since the image is reduced during template matching, the target point coordinates need to be restored to the original image. The formula is as follows:

[0091]

[0092] Where: (x, y) is the coordinate of the reduced image;

[0093] (track_x, track_y) are the coordinates of the original image;

[0094] N is half the width of the original image;

[0095] M is half the height of the original image;

[0096] r is the image scaling factor;

[0097] When the matching coefficient of the feature points is lower than a certain threshold or the relative position relationship between the feature points is deformed, the feature points need to be re-detected to avoid tracking errors. When the final target tracking coordinates are obtained by weight distribution calculation, the target template itself needs to be updated to avoid tracking drift.

[0098] Since the first two consecutive occurrences of s n <s n , that is, at this time, the missile is in the final stage of flight, and the scale of the image before and after the frame changes greatly. Directly matching the original image with the NCC template can no longer meet the accuracy requirements, so the original image needs to be reduced by a certain multiple to adapt to the image scale change. At this time, firstly, the target template Mask N The image to be matched is reduced by r times and the image reduced by r 2 The NCC template matching is performed on the image, and the matched targets are N′ n and N″ n , the corresponding matching coefficient is s′ n and s″ n The reduction coefficient r is determined by the ratio of the projectile-target distance between the current frame image and the image to be matched, and is executed in subsequent frames in a loop.

[0099] When s′ n >s″ n When the target itself is the template Mask N The matching coefficient is higher than that of the image to be matched, and then the feature point template Mask P1 , Mask P2 , Mask P3 The NCC template matching is performed with the image to be matched which is reduced by r times, and the matching feature points obtained are P′1, P′2, P′3, and the matching coefficients are Take p1 ,s p2 ,s p3 The minimum value is s pmin According to the relative position relationship between feature points P1, P2, P3 and target N, the target point N can be indirectly obtained by calculating the matching feature points P′1, P′2, P′3. P (x P ,y P ), according to the matching coefficient and pmin The coordinate point N of the matching target point indirectly obtained based on the matching feature point P (x P ,y P ) and Mask based on the target's own template N Directly obtained matching target point (x n ,y n) is used to distribute the coefficient weights, and the final target point coordinates (x, y) are as follows:

[0100]

[0101] Since the image is reduced during template matching, the target point coordinates need to be restored to the original image.

[0102] When s′ appears n <s″ n When s′ n >s″ n strategy execution;

[0103] If and only if s′ appears twice in a row n <s″ n When the target's own template Mask is used twice in a row N The reduced r to be matched 2 The image matching coefficient is higher, and then the feature point template Mask P1 , Mask P2 , Mask P3 Respectively with the reduced r to be matched 2 The NCC template matching is performed on the image, and the matching feature points obtained are P′1, P′2, P′3, and the matching coefficients are s p1 ,s p2 ,s p3 , take s p1 ,s p2 ,s p3 The minimum value is s pmin According to the relative position relationship between feature points P1, P2, P3 and target N, the target point N can be indirectly obtained by calculating the matching feature points P′1, P′2, P′3. P (x P ,y P ), according to the matching coefficient s″ n and pmin The coordinate point N of the matching target point indirectly obtained based on the matching feature point P (x P ,y P ) and Mask based on the target's own template N Directly obtained matching target point N′ n (x n ,y n ) is used to distribute the coefficient weights, and the final target point coordinates (x, y) are as follows:

[0104]

[0105] Since the image is reduced during template matching, the target point coordinates need to be restored to the original image.

[0106] Since s′ appears twice in a row n <s″ n , at this time, the same principle applies, and then the target template Mask is first N The reduced r to be matched 2 Zoom in and out 3 The NCC template matching is performed on the image, and the matched targets are N′ n and N″ n , the corresponding matching coefficient is s′ n and s″ n , the reduction coefficient r is determined by the ratio of the projectile-target distance between the current frame image and the image to be matched.

[0107] According to the above logic, whenever there are two consecutive s′ n <s″ n When , it means that the image scale changes faster, and NCC template matching can no longer meet the accuracy requirements for this scale. The image needs to be further reduced by a certain multiple to make the target template image scale and the image to be matched as consistent as possible to improve the matching accuracy. The closer the distance between the projectile and the target, the greater the reduction multiple.

[0108] like Figure 3 As shown in the figure, at the end of the flight, if only the target template is used for tracking, the target scale will change greatly at this time, and the target will soon fill the field of view, resulting in a tracking blind spot, causing the target center to deviate from the center of the field of view, and failing to meet the requirement that the projectile hits the target center. If the feature point template and the target template are used for dual matching tracking and the image is adaptively reduced, when the target template cannot meet the accuracy requirements, the feature point template can still be used to complete the target tracking because the image is reduced. Figure 3 , Figure 4 As shown in the figure, since only the target template is used for tracking, the target center deviates from the center of the field of view. At this time, the image reduction feature point template matching can still be used to correct the target center and pull the target center to the center of the field of view.

[0109] By reducing the image and performing indirect calculations through direct target matching and feature point matching, the tracking drift caused by the cumulative error of single template matching can be greatly reduced, the tracking accuracy can be improved, and finally the target tracking coordinates of the real-time image can be passed to the servo system or guidance system.

[0110] Although the content of the present invention has been described in detail through the above preferred embodiments, it should be appreciated that the above description should not be considered as a limitation of the present invention. After reading the above content, it will be apparent to those skilled in the art that various modifications and substitutions of the present invention will occur. Therefore, the protection scope of the present invention should be limited by the appended claims.

[0111] The contents not described in detail in the specification of the present invention belong to the common knowledge of the professionals in this field.

Claims

1. A multi-template collaborative optimal matching target tracking method, characterized in that: include: S1. According to the target recognition information, several target feature points are obtained through feature point detection; S2, establishing the relative position relationship between a number of target feature points and the target center point; S3, establishing several feature point templates and target templates; S4, performing NCC template matching on a plurality of feature point templates and a target template with the image to be matched, respectively, to obtain matching coefficients of a plurality of feature points and a matching coefficient of a target point; S5. Select multiple matching feature points according to the matching coefficients of the feature point templates; S6. Calculate the coordinates of the matching feature target point based on the relative position relationship and multiple matching feature points; S7, based on the target's own template, matching to obtain the coordinates of the target point itself; S8, according to the matching coefficient of the feature point and the matching coefficient of the target point, respectively assigning different weights to the matching feature target point coordinates and the own target point coordinates to obtain the tracking coordinates; S9. Update the target information according to the tracking coordinates, and repeat steps S1 to S8 until the target tracking is completed.

2. The multi-template collaborative optimal matching target tracking method according to claim 1, characterized in that: The feature point detection in step S1 is specifically as follows: Determine a feature point detection rectangular area, wherein the feature point detection rectangular area is: the coordinates of the upper left corner of the rectangle (m, m), and the coordinates of the lower right corner of the rectangle (Mm, Nm), where M and N are the width and height of the image, and m is the width and height of the feature point template; In the feature point detection rectangular area, the pixel points whose absolute difference between the pixel value and the pixel value of the surrounding area is greater than 55 are selected as feature points.

3. The multi-template collaborative optimal matching target tracking method according to claim 1, characterized in that: In step S1, the feature point detection adopts the Surf feature point detection algorithm.

4. The method for target tracking by collaborative optimal matching of multiple templates according to claim 1, characterized in that: In step S2, the relative position relationship between the plurality of target feature points and the target center point is specifically: Connect several target feature points in sequence to form an outer contour; Calculate the angle between the line connecting the target center point and each target feature point and the outer contour.

5. The method for target tracking by collaborative optimal matching of multiple templates according to claim 1, characterized in that: In step S3, the constructed feature point template and the target template are of the same size; the width and height of the template are both m, m=1.3-2.5D, and D is the diameter of the target circumscribed circle.

6. The method for target tracking by collaborative optimal matching of multiple templates according to claim 1, characterized in that: In step S4, the NCC template matching calculation formula is: Where T(x, y) is the pixel value of the template image, I(x+u, y+v) is the pixel value of the target image at position (u, v), is the average value of the template image, is the average value of the window corresponding to the target image at position (u, v), and NCC(u, v) is the NCC value.

7. The multi-template collaborative optimal matching target tracking method according to claim 1, characterized in that: In step S4, the images to be matched are the original image and the reduced image. The reduced image is the original image reduced by a certain multiple with the target point as the center. The reduction multiple is the ratio of the projectile-target distance corresponding to the template image to the projectile-target distance corresponding to the image to be matched. The reduced image is filled with pixel values ​​of 0 to the size of the original image.

8. The method for target tracking by collaborative optimal matching of multiple templates according to claim 1, characterized in that: In the step S5, a plurality of matching feature points are selected according to the matching coefficients of the plurality of feature point templates, specifically: matching feature points corresponding to the N largest matching coefficients are selected from the plurality of matching coefficients; N≥2.

9. The method for target tracking by collaborative optimal matching of multiple templates according to claim 1, characterized in that: In step S8, different weights are assigned to the matching feature target point coordinates and the own target point coordinates according to the matching coefficients of the feature points and the matching coefficients of the target points, respectively, to obtain the tracking coordinates, and the calculation formula is: Among them, S pmin is the minimum value among several feature point matching coefficients, S n is the matching coefficient of the target’s own template; (x p ,y p ) is the coordinate of the target point indirectly obtained based on the matching feature points, (x n ,y n ) is the coordinate of the matching target point directly obtained based on the target template itself, and (x, y) is the tracking coordinate.

10. The method for target tracking by collaborative optimal matching of multiple templates according to claim 9, characterized in that: When the target size is larger than the target template size, the weight of the target point coordinates is 0.

Images