RAFSet (Robust Aged Feature Set)-Based Monocular Visual Odometry
Hyunho Jeon, Jin-Hyung Kim, Yun-Ho Ko · Journal of Institute of Control Robotics and Systems · 2017
This paper proposes a monocular visual odometry algorithm based on Robust Aged Feature Set (RAFSet). Conventional visual odometry algorithms generally adopt optimization algorithms to resolve tracking and matching errors in motion estimation caused by noise or occlusion. However, continuous failures in motion estimation can lead to extreme states that cannot be resolved through this optimization algorithm. We propose the RAFSet framework that effectively manages and controls the detected features of an image sequence for motion estimation to avoid such failures in motion estimation and estimate more robust motion information. A RAFSet is composed of Robust Aged Features (RAFs) that are detected or tracked from each image frame. Each RAF has an age value representing its reliability degree as a robust feature and several pieces of useful information for motion estimation. The proposed RAFSet framework adjusts the age value of each RAF effectively and only uses RAFs with a high age value to attain better motion estimation results. Experimental results for a popular dataset confirm that the motion estimation accuracy is improved using the proposed RAFSet framework.