Ego-Motion Estimation using Normal Flows

In this paper we propose a novel method to estimate ego-motion parameters of a camera by directly using normal flows. The normal flow field is fully determined by the spatio-temporal derivatives of the image sequence. Therefore, different from traditional methods which achieve the estimation by tracking corresponding features or by calculating the optical flow field, the proposed method could achieve the ego-motion estimation without any artificial assumptions on imaging scene. Additionally, the interferences can be eliminated by the RANSAC algorithm and the reliability of the estimation is further verified by the back-tracking validation. A series of experiments with various synthetic data and real images have been conducted to test the feasibility and reliability of our method.