Research On Target Tracking With Heterogeneous Binocular Vision Based On Visible Light And Infrared Images

Computer vision research is an important branch of artificial intelligence development,and binocular vision research is an important part in the field of computer vision.The design of binocular vision is simple,the speed of algorithm is fast,and the depth information of the target object can be obtained by the parallax principle,thereby which can achieve the effect of restoring the three-dimensional information of the scene.With the development of society,researchers have paid more and more attention to the research of binocular vision,and the research of binocular vision has been widely used in various fields such as industry,medicine,security,transportation and so on.Currently,the binocular vision mainly uses a homologous sensor to acquire a sequence image,and then processes,analyzes,detects,and tracks the sequence image.However,in the complex environment with low visibility,the traditional binocular vision is difficult to obtain the information by the visible light sensor to realize the detection and tracking of the moving target.Therefore,the heterogeneous binocular vision is born.Binocular vision target tracking combined with visible light sensor and infrared sensor can make full use of the scene gray information and texture information acquired by the visible light sensor,and combine the temperature information acquired by the infrared sensor to realize information complementation in a complex environment.Continuously tracking for objectives is very important.Based on this,this paper develops a tracking study based on visible and infrared images of all-source binocular vision.The main research contents are summarized as follows:(1)Camera calibration.This paper analyzes and compares various types of camera calibration algorithms,and finally selects the Zhang Zhengyou calibration method with high flexibility and precision to realize dual-target fixation of visible and infrared image heterogeneous binocular vision cameras,and obtains camera internal and external parameters.Image correction is performed on the output image such that the output visible light is on the same horizontal line as the infrared image.(2)Moving target detection.The purpose of moving target detection is to segment the background and foreground of the image to achieve the effect of tracking the target.The current mainstream moving target detection algorithms include interframe difference method,three-frame difference method,background difference method,Vibe algorithm,Vibe+ algorithm and so on.In this paper,the advantages and disadvantages of the frame difference method and the background method arecompared and analyzed.A fast and efficient new moving target detection algorithm is proposed by combining the inter-frame difference method and the background difference method.(3)Image registration and fusion.Due to the different imaging principles of visible light and infrared images,the traditional image matching algorithm can still not meet the actual needs.According to the characteristics of visible light and infrared images,both visible and infrared images retain the scene edge information,and the morphological method is used to extract the edge information of visible light and infrared image.The improved C_SIFT algorithm is used to extract image feature points.To improve the algorithm speed,use BRIEFF The algorithm generates binary descriptors and registers them.Finally,the RASANC algorithm is used to eliminate the mismatched points to complete the registration of visible and infrared images.The registered visible and infrared images can be directly merged using image fusion algorithms,and the information complementation has been achieved.(4)Texture feature detection.In order to accurately locate a specific target location,specific target feature information needs to be extracted.LBP and its improved algorithm can generate binary descriptors to describe the target texture information,but the current LBP and its derived algorithms cannot simultaneously meet the accuracy and high speed required by the target tracking algorithm.Based on this reality,this paper proposes to improve the LBP algorithm ECLBP detection and describe the specific target texture information.(5)Target Tracking.The sequence image processed by the above steps can be directly used for the target tracking operation.Firstly,the target to be tracked is manually selected,and the specific target is tracked in real time using the MedianFlow algorithm.To improve the accuracy and effectiveness of the tracking algorithm,Kalman filter prediction is used.Target motion trajectory.