The Research Of Feature Point Extraction And Matching Tracking In Image Sequences

Feature point detection and extraction of moving object, which is detecting and extracting feature points about a moving object from 2D image sequence, and connecting all target feature points in image sequence. Finally, a continuous sequence composed by each frame feature points of moving target, which can facilitate further studies, such as 3D motion reconstruction technology required feature points. The research has always been one of hot research topics in the field of computer vision and pattern recognition, and this research is also the main part in this article.Based on the analysis of the detection and extraction technology of feature points in the image, the traditional SUSAN algorithm and Harris algorithm are studied in the paper. At the same time, the frame difference method, background difference method and optical flow method are analyzed and studied, which are used to analyze moving target tracking. Compared to other tracking algorithms, optical flow method not only contains the trajectories information of the moving target, but also contains a large amount of information of 3D structure, which can detect the moving target object when we don’t know the background or other information. However, the calculation of the optical flow method is too large, so the real-time capability is relatively weak. So the improved optical flow method is a research point of this article, which can reduce the calculation.This article mainly around the improved detection and the location prediction matching tracking of feature points to study. It contains three aspects of research as follows:(1) Aiming at some problems of traditional Harris operator, which have no scale invariance, the extracted corner is pixel level, and the detection speed is slow. In this paper, a multi-scale and at sub-pixel Harris corner detection algorithm is proposed. Through the image scale of gaussian smoothing, reuse original angular point as the center of the cluster, distance weighting on each point in the cluster, and precise positioning of the pixel. Experimental results show that compared with the traditional Harris algorithm, this proposed method can adapt to different scales and changes in the multi-scale corner detection, at the same time, the detection accuracy of angle point in reached the sub-pixel, which is more accurate than pixel.(2) The calculated amount of Optical flow method is very large, and the real-time performance is very poor. In order to solve this problem, the background difference method of optical flow is proposed in our paper. Combine the background difference method and optical flow method effectively, namely, use the background difference method for the first step of pre-processing image sequence, and then use optical flow method to calculate the results of the first step. The results show that the improved method can effectively improve the real-time performance of moving target tracking, and the system stability is good.(3) Each frame of target’s feature points in image sequence, can match the corresponding or not is also one of the research keys of this paper. The target feature point is based on the target motion trajectory of the dynamic matching of the message. The detection and extraction of the feature points of each frame image, improved optical flow method on the whole image sequence for target tracking and forecasting of image feature points, combined with each other, which can achieve the goal that detection and extraction of the feature points in image sequences.