Research On Electronic Image Stabilization Algorithm Based On Feature Trajectory Optimization

The popularization of mobile devices such as mobile phones,miniature cameras and other portable devices makes it possible to obtain videos whenever and wherever.Due to the lack of the professional camera support in shooting,the output video sequence are usually shaky,it affects the video viewing and further processing.Video stabilization becomes an important and exigent problem in video processing,which strives for eliminating the visual shake.The main content of this thesis is the electronic image stabilization algorithm which is based on feature trajectory optimization in video stabilization.The research works include:(1)We design a feature trajectory construction algorithm which is based on SURF.At first,we analyze and compare some traditional feature point extraction algorithms,and select the SURF algorithm to extract feature points.Next,we analyze and summarize the shortcomings of conventional K distance ratio method,and propose a new method based on double threshold constraint and Bi direction detection for K distance ratio method to improve the accuracy of feature point matching.Then,we propose a RANSAC algorithm which is based on minimum distance deviation and block processing,the algorithm not only improves the accuracy of matching pairs,but can also deal with multi-plane images.Finally,we design a robust feature trajectory construction algorithm on the basis of the above research.(2)We design a motion optimization algorithm based on feature trajectory optimization.First of all,we analyze and conclude the shortcomings of predecessors,then propose an optimization scheme that combines the motion smoothness and motion compensation.Then,we conduct the researches for the constraints in the trajectory smooth and space,and propose a whole motion optimization scheme.(3)We design an electronic image stabilization system which is based on feature trajectory optimization.Firstly,on the basis of the feature trajectory construction algorithm and motion optimization algorithm,we utilize the thought of segment handling to design the stabilization image system,in this way,we enhance the system execution efficiency.Then,we divide the system into several segments,and conduct the class design.In the end,we compare our algorithm with other state of art algorithms.Theoretical analysis and experimental results show that the research results of this paper solve the jitter problem in the sequence of video sequences to a certain extent.