Vehicle Ego-motion Estimation using Optical Flow and Fundamental Matrix Evaluation

Abstract

Estimating precise vehicle ego-motion is a key task for the safety of drivers in advanced driver assistance systems (ADAS). In this thesis, we introduce a method for the ego-motion estimation based on an optical flow algorithm and the fundamental matrix. Since the resulting fundamental matrix has a critical role in the stability of the final result, we evaluate the fundamental matrix using two optimization algorithms and verified that the proposed algorithm improves the accuracy and stability from the initial estimation. Also, in the estimation of the ego-motion of a vehicle, the velocity of the vehicle acts as an important parameter affecting the stability of the fundamental matrix. For instance, the slower velocity of the vehicle, the higher state instability. Therefore, in this thesis, we also introduce and verify that merging optical flow information improves the stability of the fundamental matrix by comparing the directions of the vehicle when the vehicle is in slow transitional motion. Last but not least, we demonstrate that the stability of the fundamental matrix is greatly improved by bundle adjustment that optimizes the total landmarks and the ego-motions in several frames simultaneously.