Photoelectric detection air target matching method for point target

Abstract

The present application relates to a kind of photoelectric detection aerial target matching method for point target, for several frames of multi-target image obtained by distributed imaging system, traverse target coding and spatial coordinates, multi-target full path length is solved by synthesizing above information, find the shortest path using shortest path convex optimization algorithm, to realize the matching of multiple targets in different detectors, then form the corresponding relationship of the observation of the same target by multiple detectors, and effectively realize the cross positioning function to the same target.The present application can be applied to distributed photoelectric cooperative detection multi-target cross positioning.Compared with prior art, the beneficial effects of the present application are: for multiple aerial point targets in the same field of view, the target corresponding relationship can be formed faster, suitable for multi-target tracking scene;It is beneficial to reduce the false points generated by direction-finding line intersection, improve the accuracy of multi-target matching, and reduce the false alarm rate.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of binocular vision principle, camera distributed multi-frame shooting, ranging positioning, target matching under convex optimization algorithm, and particularly relates to a method for matching point target aerial targets by using distributed photoelectric detectors. BACKGROUND

[0002] When the conventional photoelectric detection equipment detects point targets, the matching ability of different photoelectric detectors for point targets is poor when there are multiple targets in the field of view, and the cross-positioning ability for the same target cannot be timely and accurately formed, resulting in too many false points in cross-positioning, poor cross-positioning accuracy and other phenomena.

[0003] In actual engineering, two basic methods are often used for cross-positioning: one is a long-term observation method for targets, which selects target 1 from the first detector, selects suspected matching target 2 from the second detector, and then determines whether target 2 is the same target as target 1 by observing the movement and radiation rules of target 1 and target 2 for a long time. If not, traverse other targets in the second detector until a target similar to target 1 is found, and form the matching target of target 1, and then perform target cross-positioning by solving the direction-finding line. Since the tracking time is generally long, this method is usually suitable for single target tracking and positioning; the second method is to solve the direction-finding line for each target in the field of view, and then perform cross-positioning. For N targets, the number of cross lines is N, and the number of intersection points is N 2 . 2 -N false points are excluded one by one. SUMMARY

[0004] In view of the defects in the prior art, the present application provides a point target matching method, which can be used for the cooperative detection technology and multi-target cross-positioning technology of distributed photoelectric detection equipment, and is beneficial to eliminate false points in cross-positioning, and realizes accurate matching and cross-positioning of photoelectric detection equipment for aerial point targets.

[0005] The present application provides a photoelectric detection aerial target matching method for point targets, comprising the following steps:

[0006] Step 1: using 2 cameras to shoot n groups of unmanned aerial vehicles, a total of M frames, wherein the shooting time between frames is extremely short, and the unmanned aerial vehicle targets in different frames shot by the same camera can be distinguished;

[0007] Step 2: processing the same frame shot by the two cameras, and numbering the key points obtained by camera 1 and camera 2 in the first frame as and Where k1 is the number of targets observed by camera 1, and k2 is the number of targets observed by camera 2, and i takes 1, 2, …, k1 or k2, which are not corresponding matching points;

[0008] Step 3, extend the numbering method of 2 cameras to p cameras, where p≥2, and the target number observed by the first p camera in the first frame is:

[0009] Step 4, for multiple cameras, arrange and combine multiple target matching schemes, and set the number of matching schemes as A. When p=2, k1=k2=n, and each target is imaged in the camera, there are A=n! matching schemes in total.

[0010] Step 5, for the i-th scheme, the matching points are 1, 2, 3, …, B i , B i represents all matching target points in the i-th scheme.

[0011] Calculate the moving distance of the target in M frames of image point by point: where d i,j is the moving distance of the j-th target in the i-th scheme in M frames of image, is the position vector of each point, j=1, 2, …, B i ; q=1, 2, …, M, the upper right corner number q represents the q-th frame, and the vector and the vector are the position vectors of the same prediction point in different frames.

[0012] Step 6, calculate D i , which represents the sum of the moving distances of all target points in the i-th scheme.

[0013] Step 7, find the minimum value in the array {D i}, D min =min({D i}), at this time the corresponding target matching scheme is the optimal matching scheme.

[0014] Step 8, if other target matching schemes are close to D min , it means that part of the matching points are not sensitive to the matching strategy, and these matching points are not successful, and need to be observed for a long time and execute steps 1-7.

[0015] Step 9, continue to observe the target, use the above steps to match and check, correct the matching result, and cross-position the convergence point of the matching result.

[0016] Compared with the prior art, the present application has the beneficial effects that: for multiple aerial point targets in the same field of view, the target correspondence can be formed faster, which is suitable for multi-target tracking scenes; false points generated by the intersection of direction-finding lines are reduced, the multi-target matching accuracy is improved, and the false alarm rate is reduced. BRIEF DESCRIPTION OF DRAWINGS

[0017] BRIEF DESCRIPTION OF DRAWINGS Figure 1 is a schematic diagram of a target scene;

[0018] BRIEF DESCRIPTION OF DRAWINGS Figure 2 is a flowchart of a target matching method. DETAILED DESCRIPTION

[0019] The specific embodiments of the present application will be described in detail below with reference to the accompanying drawings, but it should be understood that the protection scope of the present application is not limited by the specific embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor are within the scope of protection of the present application.

[0020] The present application is applied to a distributed photoelectric detection system composed of multiple cameras and multiple frames, and the specific scene is as shown in Figure 1 , which specifically includes n cameras and multiple unmanned aerial vehicles distributed and measured, and the enlarged view is a single frame image taken by each camera from different positions.

[0021] Referring to Figure 2 , based on the above target scene, the specific technical implementation steps of the present application are:

[0022] ①Suppose that 2 cameras are used to take pictures of n unmanned aerial vehicle groups, a total of M frames are taken, and the shooting time between frames is extremely short, so the unmanned aerial vehicle targets taken by the same camera between different frames are considered to be distinguishable.

[0023] ②The same frame picture taken by the 2 cameras is processed, at this time, due to the problems of possible occlusion and long distance between camera distribution, it is difficult to calibrate multiple cameras and realize point matching between different cameras for n unmanned aerial vehicles by using existing stereo matching methods. Therefore, the key points obtained by camera 1 and camera 2 in the first frame are numbered, respectively and , wherein k1 is the number of targets observed by camera 1, k2 is the number of targets observed by camera 2, and k1 and k2 are generally not greater than n, and may be greater than n due to other interference. It should be noted that and are not corresponding matching points, i is 1, 2, …, k1 or k2.

[0024] ③Generalize the numbering method of two cameras to p cameras, and the number of cameras p≥2. The target number observed by the first frame of the pth camera is:

[0025] ④For multiple cameras, arrange and combine multiple target matching schemes, and the number of matching schemes is A. When p=2, k1=k2=n, and each target is imaged in the camera, there are A=n! matching schemes.

[0026] ⑤For the ith scheme, the matching points are 1, 2, 3, …, B i , B i represent all matched target points in the ith scheme; calculate the moving distance of the target in M frames of image point by point: where d i,j is the distance moved by the jth target in the ith scheme in M frames of image, is the position vector of each point, j=1, 2, …, B i ; q=1, 2, …, M, the upper right corner number q represents the qth frame. Because the time between different frames is very short, it can be considered that when the right lower corner numbers of the predicted points are the same, the two points are the same predicted point obtained in different frames, i.e. vector and vector are the position vectors of the same predicted point in different frames.

[0027] ⑥Calculate D i , which represents the sum of the distances moved by all target points in the ith scheme.

[0028] ⑦Find the minimum value in the array {D i}, D min =min({D i}), at this time the corresponding target matching scheme is the optimal matching scheme.

[0029] ⑧If other target matching schemes are close to D min , it means that part of the matching points are not sensitive to the matching strategy, and these matching points are not matched successfully, and need to continue to be observed for a long time and execute the processes of ①-⑦.

[0030] ⑨Continue to observe the target, use the above steps to match and check, correct the matching result, and cross-position the convergence point of the matching result.

[0031] The present application proposes a multi-target matching recognition method of the shortest path convex optimization algorithm, which can be applied to distributed photoelectric cooperative detection multi-target cross positioning. The core idea of the algorithm is to traverse the target coding and spatial coordinates for several frames of multi-target images obtained by the distributed imaging system, to calculate the total path length of the multi-targets by comprehensively using the above information, to further obtain the shortest path by combining the convex optimization algorithm, to realize the matching of multiple targets in different detectors, and then to form the corresponding relationship of the observation of the same target by multiple detectors, and to effectively realize the cross positioning function of the same target. The technology can also solve the effective recognition and matching positioning of individual targets under the condition of occlusion.

[0032] Taking multiple unmanned aerial vehicle targets as an example (which can be extended to other aerial targets), the present application uses multiple cameras as detectors, and the multiple cameras cross detect the unmanned aerial vehicles. First, the targets in the different frames of images of the two cameras are numbered. Then, the targets detected by different cameras are arranged and combined to obtain the relevant position vectors. Then, the total distance of the movement of all targets under different frames is calculated, and the minimum value of the movement distance is searched by using an optimization algorithm to realize the positioning and matching of the targets.

[0033] The above embodiments are only preferred embodiments of the present application, so equivalent changes or modifications made according to the structure, features and principles described in the patent application range of the present application are included in the patent application range of the present application.

Claims

1. A method for matching optically detected aerial targets to point targets, characterized by, Comprising the following steps: Step 1, using 2 cameras to take pictures of n groups of UAVs, a total of M frames, wherein the shooting time between frames is extremely short, and the UAV targets between different frames taken by the same camera can be considered distinguishable; Step 2, process the same frame picture taken by two cameras, number the key points obtained by camera 1 and camera 2 first frame respectively as and Wherein k1 is the number of targets observed by camera 1, wherein k2 is the number of targets observed by camera 2, and Not corresponding matching points, i takes 1, 2…k1 or k2; Step 3, extend the numbering method of 2 cameras to p cameras, the number of cameras p≥2, then the target number observed by the first frame of the pth camera is: Step 4, for multiple cameras, arrange and combine multiple target matching schemes, set the number of matching schemes as A, when p=2, k1=k2=n and each target is imaged in the camera, there are A=n! matching schemes in total; Step 5, for the i-th scheme, the matching points are 1, 2, 3, …, B respectively i , B i denotes all the matching target points in the i-th scheme; The target moving distance of M-frame images is calculated point by point: Wherein d i,j is the distance of the jth target in the ith scheme moving in the M-frame image, is the position vector of each point, j = 1, 2, …, B i ; q = 1, 2, …, M, the upper right corner number q represents the qth frame, and the vector is the position vector of the same prediction point in different frames; and is the position vector of the same prediction point in different frames. Step 6, calculation D i denotes the sum of the distances of all target points moving in the i-th solution. Step 7, find the minimum value in the array {D i}, D min = min({D i}), at this time the corresponding target matching scheme is the optimal matching scheme; Step 8, if other target matching schemes match with D min The proximity indicates that the partial matching point pair matching strategy is not sensitive, and these matching points are not successfully matched, and long-term observation needs to be continued, and steps 1 to 7 are executed. Step 9, continuously observe the target, use the above steps to match and check, correct the matching result, and cross locate the convergence point of the matching result.

Images