结合改进的SIFT特征匹配方法的运动跟踪算法-控制理论与控制工程专业论文.docxVIP

上海交通大学硕士学位论文结合改进的 上海交通大学硕士学位论文 结合改进的 SIFT 特征匹配方法的运动跟踪算法 II II The Motion Tracking Algorithm combined with the Improved SIF T Method Abstract Motion target tracking is always an important research subject in Pattern Recognition. It can be widely used in video surveillance, human-computer interaction, intelligent environments, video compression and military field etc. The main difficulty with the algorithm is how to effectively deal with various situations in the real complex environment. At present, there is still not an effective method to cope with object occlusion, angle variation, scale change, non-rigid object etc. Mean-Shift is a fast algorithm for motion tracking based on color of the object. But it cannot effectively track object when the color is similar between the object and environment in that it tracks object on the basis of color. And it is also difficult for the algorithm to deal with the objects of non-pure color and occlusion. To solve the problem mentioned above, the paper presents a new motion tracking method, which combines the improved SIFT feature matching with Mean-Shift for the first time. It uses the SIFT method to extract features in combination with the color to perform the reliable tracking. The algorithm in the paper uses rectangle operator to simulate the gaussian convolution and proposes the reverse rules to determine the extreme points, which accelerates the detection of extreme points and effectively improves the speed of SIFT algorithm. The paper also proposes a new improved particle filter in which part of the particles are randomly oriented. The new method III III is aimed to deal with the sample depletion problem in traditional particle filter. The improved particle filter algorithm is testified through experiment to greatly reduce the error caused by the sample depletion. The whole tracking algorithm uses feature points extracted by improved SIFT as the reference points of Mean-Shift and calculates the target area center, which