Research On Integrated Tracking Algorithm Based On Fisheye Lens And Implementation In Smart Camera

A lot of research has been done on fisheye lens which has been used widely in many application areas due to it super wide view angle in machine vision field. In this paper, recognition and tracking algorithm has been studied based on fisheye lens vision system, and a success experiment has been done to verify the improved algorithm.Image tracking developed very fast, expert both in internal and external has spent a lot of time and energy on it and obtain a lot of useful achievement. Particles filter and mean shift which are all relate with probability theory, is the most popular tracking algorithm, both them are widely used and researched. They have different advantages and disadvantages. Particles filter is applicable in non-gauss and non-linear, have high robustness to interference noise and non sensitive to overlap. But more accuracy it tracking, more particles will be needed. So it will spend more time to process one frame. Mean shift has opposite merit, only a few iterations is needed to track on frame, that is much less time than particles filter.Correspondingly, when a interferent show up or the tracked object missing immediately, the algorithm will failed.Two improved tracking algorithm applied in different environment are presented in this paper after studied on the two methods. In outdoor environment, algorithm always fails on tracking object using only color feature. So, multi-features combination is proposed to cover the defect of color based algorithm. We use democratic strategy to integrate the different feature. Multi-features based particles filter performs very well on tracking street lamp in outdoor environment.In another improved tracking algorithm, we combined particles filter and mean shift, keep advantages of them but no disadvantage. The new algorithm still be the same with particles filter applied in non-gauss, non-linear, robustness and non sensitive to interfere, but more efficient than classical particles filter, like mean shift. In the end, a contradistinction between the classical tracking algorithm and the improved methods are given, and the experimental result shows that the improved algorithm are more efficient and robustness than the old one.A special-designed embedded system called BlueEye smart camera which has the advantage of small volume, low power consumption and high computation capability is designed in the end of the paper. The BlueEye smart camera integrates CMOS, DSP and FPGA into one system, has the ability of image capture and process. After implement the improved tracking algorithm in BlueEye smart camera, we also developed the client program executing in the PC. The experimental result shows that both the algorithm and the smart camera are perform well.