Design and Evaluation of Compact Real-time Descriptor for Camera Motion Estimation

Abstract

This paper proposes the image processing algorithm of camera motion estimation that can be used for autonomous mobile robots. Most autonomous mobile robots need to quickly recognize the surroundings so as to move to the space. Moreover, the image processing algorithm requires low computation complexity because the mobile systems need to fulfill the immediate reaction for the spatial movement. Therefore, Compact and Real-time Descriptors (CARD) was proposed in order to extract local image features for corresponding to keypoints between two images. Based on CARD, the proposed image feature descriptor has the ability of scale-invariance and rotation-invariance for extracting image features and is represented by binary description so as to use small memory capacity. Our method uses a multi-scale image pyramid to extract coordinates for keypoints and a log-polar binning pattern for a patch around the keypoints. The results of correspondence for keypoints between two images were good enough to estimate the camera motion, where comparisons to other descriptors such as SIFT, SURF, and FREAK were evaluated.