Research on 3D reconstruction for robot based on SIFT feature

Abstract

On the basis of only visual and odometer, a robust perception model is established to extract environmental features through effective fixed scale feature-transformation method, and updated feature by unscented Kalman filtering. The scale invariant feature transform (SIFT) is studied for 3D reconstruction, and a fast feature matching algorithm based on SIFT is proposed. A map representation method using SIFT features is also propounded, which is more convenient for environment recognition, robot localization and makes the data association map building much easier as well than the maps using simple features such as Harris corners and edges. The results of experiment show that this method can improve the success rate and precision of robot localization.