Research On Algorithm Of Moving Object Detection And Tracking In Video

Video moving target detection and tracking technology is a hot research topic in the field of computer vision.It is widely used in intelligent surveillance,human-computer interaction,unmanned reconnaissance aircraft and other fields.In recent years,with the development of filter tracking algorithm and the rise of deep learning,a variety of new moving target detec-tion and tracking technologies have emerged,but these algorithms are highly complex and difficult to be large-scale applied in engineering.In this thesis,the radar and photoelectric joint monitoring and control system for photoelectric the moving target detection and track-ing problems are studied.In order to solve the problems of remote imaging blur,ghosting,occlusion and rotation of the target in the process of detecting and tracking moving targets in desert,Gobi and other practical applications,a moving target detection and tracking system matching the working mode of photoelectric terminal was designed.The system is divided into two parts:moving target detection and target tracking.The two parts work together to continuously and steadily track the moving target and apply it to practical engineering deployment.The main contributions can be concluded as follows:In terms of moving target detection,considering the frequent transfer of the radar to the photoelectric system in the radar and photoelectric joint monitoring system,the fast moving target detection method is adopted to capture the target.In addition,as the moving target in this scene is too small,which only takes up hundreds of pixels,we select the background modeling method of ViBe(Visual Background Extractor)as the base algorithm.In this thesis,the method of extracting the moving region and modeling part of the background is adopted to solve the problem of excessive calculation amount of background modeling algorithm caused by excessive image resolution.At the same time,to improve the ViBe background modeling algorithm,we propose a dual background modeling method,which can quickly filter the ghost in moving target detection.In addition,a scintillation point filtering algorithm is proposed to filter scintillation points in the background,which enhances the adaptability of the algorithm to rain and snow weather.On the basis of target detection,considering the harsh environment,difficult conditions and limited performance of the terminal in the actual application of the system,we employed the Kernelized Correlation Filters algorithm as the base of target tracking algorithm.At the same time,an adaptive target block tracking algorithm is proposed to deal with the partial occlusion effect caused by trees,mountains and potholes in desert areas.In addition,consid-ering the characteristics of the large difference between the target color and the surrounding environment color in the application scenario of the system,a color space confidence map algorithm is proposed to improve the accuracy of the target tracking algorithm.By com-paring the simulation performance of the proposed algorithm with that of existing tracking algorithms on database OTB 100,VOT 16,VOT 19 and VisDrone 18,the effectiveness of the proposed algorithm is verified.In the last,the moving target detection and tracking system designed in this thesis is imple-mented in an engineering environment.The results of on-site video experiments verify the engineering feasibility of the algorithm proposed in this thesis,providing theoretical basis and data reference for subsequent large-scale project deployment.