OFVO: A Visual Odometry Designed for Motion Trajectory Estimation of Autonomous Vehicles

Abstract

Camera motion estimation is crucial for measuring the trajectory of moving objects. Fortunately, there is an odometry to help us solve this problem. Unlike other research works that primarily rely on odometry for indoor robots and drones, we aim to utilize visual odometry to enhance our autonomous driving perception system. The outstanding contributions of this paper are as follows: Firstly, by utilizing the feature point method in image processing, we explore various feature extraction and matching algorithms to develop a visual odometry system based on Oriented FAST and Rotated BRIEF (ORB) and Fast Library for Approximate Nearest Neighbors (FLANN) for autonomous cars, which is named the ORB-FLANN Visual Odometry System (OFVO). Secondly, we validate our algorithm in the self-driving simulator Carla and achieve satisfactory results.