IMU-aided SURF feature matching for relative pose estimation

摘要

Vision based navigation systems for robotic applications often make use of corresponding feature points between image frames. For robust relative pose estimation between captured images, it is necessary to extract distinctive features, typically with large descriptors, leading to slow feature matching. Feature matching is always subject to false matches, reducing the efficiency of relative pose estimation. In this paper we propose a method for improving the speed and outlier rejection of the feature matching step by making use of information from an Inertial Measurement Unit (IMU). The use of an approximate relative camera pose provided by the IMU is shown to reduce the search space for each feature correspondence, improving computation time, and reducing incorrect feature matches.