Research on Moving Object Detection and Tracking in Complex Condition
|School||Nanjing University of Aeronautics and Astronautics|
|Course||Applied Computer Technology|
|Keywords||Moving target detection and tracking Multi - feature fusion modeling Particle filter Line scan DSP transplant Optimization|
Moving target detection and tracking is a key topic in the computer vision -disciplinary research . Has a wide range of applications such as video surveillance, security suspects under strict surveillance , weaponry . Through the joint efforts of the researchers at home and abroad over the years , moving target detection and tracking technology has been developed by leaps and bounds , and made ??a lot of ground - breaking research . But in actual complex environment of the scene , the illumination changes , object occlusion , the presence of shadow interference , deformation effects , presented new challenges to the technology . To solve these problems well , and to achieve the requirements of practical applications , the need to design real-time and robustness of the new algorithm both . This paper in-depth study of certain core algorithm , and made ??a new attempt on this basis , this paper is as follows : depth study of the classical methods of detection and tracking of moving targets at home and abroad , and experiments to understand the advantages and disadvantages of each and suitable application conditions. On this basis , the proposed combined modeling based on multi-feature target Mean-Shift and Particle Filter tracking algorithm . The goal modeling combines general target characteristics as color , edge , texture , and also draw on the concept of structural similarity (SSIM) image quality evaluation in the field of research , the structural information extracted target characteristics as the fourth fusants . Improved multi- feature fusion method , the robustness of the target model . Final tracking of moving targets in MSPF (Mean-Shift Particle Filter) improved tracking framework . The algorithm to track accurately verified by experiments . Proposed a target extraction algorithm based on line scan , the algorithm for progressive scan image , extracted directly target chain code tracking , do not need the full map . The algorithm combined in the development of embedded DSP DMA double buffering technology , can greatly reduce the overhead of algorithm time . The multi - feature fusion the MSPF target tracking algorithm transplant to the the ADSP BF561 development platform , a number of issues to solve transplantation in image format conversion . Combined target extraction algorithm based on line scan , optimize the realization of the target tracking algorithm in the platform . Transplant based on the C language and assembly-level code optimization , to further enhance the efficiency of algorithm execution , and ultimately to meet the requirements of real-time target tracking applications .