An Adaptive Uniform Distribution ORB Based on Improved Quadtree

Abstract

ORB (Oriented FAST and Rotated BRIEF) feature is wildly applied in visual SLAM because of its excellent computational efficiency and stability. Aiming at the problem of uneven distribution of ORB feature, and improving the calculate efficiency of feature extraction at the same time, we proposed an ORB feature extraction algorithm based on improved quadtree in this paper. The proposed algorithm will select the threshold adaptively for FAST extraction according to the gray image instead of the value set artificially. And then we set different depth of quadtree according to the expected feature number which decreases as the number of image pyramid layers increases to reduce redundancy. The remained key points selected by Harris score will distribute well in the image. The results show that the proposed algorithm can improve the uniformity of ORB feature, and reduce feature extraction time compared to the algorithm in ORB_SLAM, it has certain application value for the realization of real-time SLAM system.