Video Synchronization Based On Local Homography Matrix

Video synchronization is a main problem in the field of computer vision,which is a technique that temporally and spatially aligns unsynchronized video sequences by uncalibrated cameras from distinct viewpoints or over different time intervals.The purpose of video synchronization is to establish the relationship between time and space among the input video sequences.And it has been widely used in both military and civilian fields.Having studied the disadvantage of existing video synchronization algorithms based on the previous research work,a video synchronization algorithm based on local homography matrix is proposed.Finally,several sets of experiments are performed to demonstrate the validity of the proposed algorithm.The main contents of this dissertation are summarized as follows.First,some existing video synchronization methods are summarized.Especially four feature-based algorithms are discussed in detail in this dissertation,including sparse projective invariant representation,motion-based,time line constraint and rank constraint.Experiments show that these feature-based video synchronization algorithms have higher alignment error for non-planar scenes.Then,a video synchronization method based on local homography matrix is proposed to overcome the following limitations.Most existing methods have higher video synchronization error and even fail for non-planar scenes.The proposed video synchronization method consists of the following parts:(1)Extract the trajectory points and the background images of the reference video and the video to be synchronized;(2)Obtain the matched feature point pairs between background images to estimate the epipolar geometry information;(3)Based on the video background image information and the epipolar geometry information,a condition of the motion trajectory matched point pairs based on the local homography matrix is obtained.Then the initial matched trajectory point pairs between the reference video and the video to be synchronized are also obtained;(4)Refine the final trajectory matched point pairs by using the epipolar geometry constraint;(5)Extract the frame indices of the matched trajectory point pairs and obtain the matched frame pairs.The temporal parameters between video sequences are estimated by using these matched frames.Finally,several pairs of videos,including synthetic scenes and real scenes,are employed to demonstrate the validity of the proposed method.Experimental results demonstrate that the proposed technique provides a better performance for planar scenes as well as for non-planar scenes than some existing methods.