Research On Human Target Tracking Technology Of Visible Light Based On Mav

In recent years, human target tracking technology based on visual light is rapidly developed and widely used in military and civilian fields, such as anti-terror work, video surveillance in public places and disaster relief. The realization of human target tracking system with full or partial intelligence can drastically diminish man laboring hours and working strength.The micro UAV human target tracking system based on image, hanged under the MUAV, is used to watch and track the moving objects on the ground, with the result that the position information of the objects in the images can be obtained. Tracking in such scenario has following characteristics: video images are blured, and contain lots of noise; Objects areas in the image are so small that less information can be obtained; The background and human targets in the image are respectively with global movement and local independent movement; At the same time, some changes can occur at any time in the scene. Therefore, the key problems of target tracking technologies can be viewed as how to select appropriate features to describe the target and what efficient target searching and matching algorithms can be chosen to ensure real-time and accurate tracking.In order to adapt to the tracking of moving targets on the ground in the scene of aerial video image, the paper delves into current object tracking algorithms, especially focuses on the research and analysis Mean Shift algorithm. At last, it is proved by experiments that the algorithm has the advantages of good real-time performance and high robustness.But it only utilizes single feature to describe the target, which is hard to accommodate complexity aerial photography scene.So in this paper a target tracking algorithm based on Mean Shift integrated color and edge features is proposed. First, kernel histogram probabilistic models is used to describe object features, and weighting fusion of Bhattacharyya is employed to evaluate the similarity of the models and the candidates in the tracking, and a selective template updating method is given according to the adaptive weights, which can overcome model shift in the tracking process.In the last, a target tracking system is designed for aerial filming scene in this paper. First the system makes use of Gaussian filter to remove image noise caused by high altitude shooting; then the target is detected manually; finally the improved algorithm proposed by this paper is employed to track the target.Experimental results demostrate that under the circumstance of camera moving, the tracking system designed in this paper has better real-time performance, accuracy and robustness to such situation as target transformation and rotation, illumination changes in the environment, target partial occlusion and so on.