Research Of The Single Object Tracking Method For The Image Sequences

Vision is a main approach to get the external information for human beings, and machine vision systems are used to recognition, measurement and tracking in stead of human beings. The tracking problem of moving object is now a hotspot in the study of machine vision. Recently, with the upgrade of cutting-edge technology and industrial level, object tracking technology is becoming riper, and has been widely used in community monitoring, defense security, and automatic production, etc. How to build a robust object model and search target is the emphasis of the research.Conventional object models take only one kind of feature information such as intensity, color, texture, contour feature into consideration. They all have their own limitations, therefore, they can hardly achieve robust tracking in the complex environment. As a result, a multi-feature fusion method is used to make the best of information and build a robust model. In order to deal with the nonlinear and non-Gaussian filter problems, a target searching method based on particle filter is used. The primary works in this thesis are listed as follows:Firstly, the fundamental theory of particle filter is introduced, and multi-dimensional object state model, motion model and observation model are built. What’s more, the framework of object tracking method based on particle filter is given.Secondly, in order to deal with the motion blur and optical fuzzy in the image sequences, color information is adopted and the object model based on color histogram is built. Compared to traditional methods, the two dimension Gauss surface template is used to allocate the weight of each pixel according to its distribution, therefore it improves the stability of tracking and reduces the operation time.Thirdly, in order to deal with the affine transformations and partial occlusion in the image sequences, the object model based on Scale Invariant Feature Transform (SIFT) feature is built. A correct method is proposed to reduce the error caused by concentrated distribution of the feature points.In addition, an algorithm based on fuzzy math is proposed to fuse the color feature with the SIFT feature and build a robust object model.Finally, video images acquired in different scenarios are used in the experiments, and the results show that the proposed method is feasible and robust.