Abstract: In target positioning system, which is based on the matching of local characteristic points, the proportion of false matching points directly impacts positioning accuracy. In traditional SIFT method, in order to guarantee there are enough matching points during the matching process, usually a minor statistic field is selected. But this results in false matching because it loses relevance of surrounding fields. In order to improve the accuracy of positioning, presents an improved SIFT characteristic descriptor, which introduces PCA statistic characteristics into traditional SIFT local descriptors. In this way, a higher correct matching ratio is obtained. Traditional filtering algorithm of SIFT false matching (e.g.: NN method) loses too much correct matching while filtering out the false ones. To further reduce the false matches, proposes a second filtering which is based on the NN method, using the matching points’ coordinate information. Experiments show that this positioning method’s accuracy is better than traditional ones.

Key words: Target , location, Feature matching, Principal component analysis