Study on Improving Image Feature Points Detection and Matching Accuracy in Binocular Vision System

Image feature points detection and matching is the key to binocular vision system performance. The paper is to improve its matching accuracy. In the experiments, Harris algorithm, Susan algorithm and CSS algorithm were used on the same image to extract feature points. Compared with each other, three methods showed different advantages in terms of extracting feature points. And two methods were carried out in the feature points matching process, one method was based on Harris feature points detection while another method was based on SIFT algorithm. The results showed that SIFT algorithm had better matching effect, but matching accuracy remained to be further improved. As a result, we extended the search scope of the extreme points in DoG scale space of the SIFT algorithm and removed feature points around image boundary. Though the number of the detected points changed little, but its detecting accuracy was more reliable. Compared with the effect of traditional SIFT algorithm, the matching accuracy has been significantly improved.