A comparison of local linear feature extraction

Abstract

A number of studies have been carried out on the passive ranging method based on the monocular imaging features to non-cooperative target. This paper deals with target ranging estimation system focus on image linear feature. The ranging system implements target ranging by means of adjacent image matching method to extract target feature points, then obtain target rotational invariant linear feature, and combined target azimuth and pitching relative to camera when image is taken with camera space coordinate, the target distance to image pickup system is gained by solving the certain target ranging equation. As for target linear feature extraction, the paper applies three algorithms of the sub-pixel Harris corners method, the Simplified Scale Invariant Feature Transform (SSIFT) method and Speeded Up Robust features (SURF) method to extract linear feature and makes an analysis to ranging performance. It implied by our experiment that the SURF algorithm is the best one in the three methods. Its computational error is relatively small, and the time consumed is shorter compared with other two algorithms. The error of the sub-pixel Harris algorithm is a bit of larger than SSIFT algorithm while the real time realization performance is better than SSIFT algorithm. © 2012 Springer-Verlag GmbH.