Tracking Algorithm Dm642 Based Camshift Goals

Target tracking technology is a hot topic in the field of computer vision, whether in civilian or military, having a profound theoretical significance and a broad application prospects. But if we want to achieve wide application of high precision in real time tracking, there is a long way to go. At present, it's difficult to track target with achieving high robustness and real time especially under complex background.Among many tracking algorithms, this article studies camshift algorithm based on kalman filter and camshaft based on inter-frame difference, then compares their effects in practical application, finally puts forward a new camshift algorithwm based on background weighed histogram and inter-frame difference, and verify it on the SEED VPM642 evaluation board. Camshift, which is based on the Meanshift, uses color probability distributions of the target to track.the algorithm not only inherited meanshift simple and real time characteristics, and can get good tracking effect with deformation occuring. Camshift algorithm based on kalman filter has very good effect at the fast-moving target and the mask question, but it is semi-automatic tracking, needs the artificial localization, poor anti-interference ability in the background of similar color with targets. Camshift algorithm based on inter-frame difference is a full automatic target trackong algorithm, but it is easy to lose the target under the complex background. completely blocked, target tracking fast motion. Finally propose a new advanced camshift target tracking algorithm, which is based on background weighed histogram and inter-frame difference. First calculate the background weighed histogram, then use inter-frame difference method automatic definiting target sector, during running the camshift algorithm process, use inter-frame difference method compareing the H component mean value and the target template datum value size, thus real-time updating target template, reduces the illumination change influence; The real-time comparing each search window and the initialization window size reduces the close color disturbance in the background. Also in the tracking process, use motion estimation to prevent rapid movement when the moving target tracking failure.Through a series of measures, the improved camshift algorithm not only has good real-time ability as the traditional camshift algorithm, but also maintains good robustness and accuracy in the complex background.