Visual loop-closure detection method using average feature descriptors

Abstract

This paper proposes a novel visual loop-closure method using average feature descriptors. The average feature descriptors are computed by averaging the descriptors of feature points at each frame. Through GPGPU (General-Purpose computing on Graphics Processing Units) technique, the proposed method selects a frame having a minimum distance with the current average feature descriptor from the average feature descriptor history. After the minimum distance calculation, loop-closure is determined by matching feature points between the selected frame and current frame. Experiments results demonstrate that the proposed method successfully detects the visual loop-closure and is much faster than the conventional visual loop-closure method in detecting the visual loop-closure.