Research And Application Of Image Registration Based On Video Sequence

With the rapid advances in digital technology, the analysis and processing of video sequence has gained increasing attention in recent years. Digital image sequence can be obtained by optics device, such as camera, which is a series of projection from the real world to imaging plane at different times. Image frames have great correlation and information redundancy. The key of research is to find and describe the relationship between images. Image registration techniques can effectively solve such problem.Image registration is a basic problem in image processing. It matches two or more images, at different times, from different sensors or from different view points, in order to remove of the potential geometrical distortion existed. Image registration is extensively applied to computer vision, pattern recognition, medical image processing and remote sensing information processing.This paper presents a new method of image registration based on adaptive clustering, which satisfy vast image registration and work fast. The algorithm clusters the matching points adaptive, based on the vector consistency between correct matching points. The experimental results show that the adaptive clustering algorithm can work reliably, filter candidate matches well, prepare for RANSAC matching and increase the speed of match processing.This paper presents a simple and effective method of image composition. For the image sequences which are obtained by fixed, horizontal panning camera, The method chooses some key frames from image sequences first, then find the overlap areas by homograghy matrix, composite the image by linear interpolation. If panning angle of camera is less than180 , the method work well.This paper measures the data of athlete successfully using image registration technology. We estimate the camera motion parameters, in order to remove of the influence of camera motion, then we can obtain real data of athlete. According to this principle, we establish a reasonable model for ice stadium, and give a corresponding method of camera shoot, design athlete location algorithm, estimate the camera panning angle by homograghy matrix of adjacent images, then we can measure athlete data by estimating the rotation angle of athlete.