Ego-motion estimation using sparse SURF flow in monocular vision systems

Abstract

The multilayer bucketing screener has proved to be effective to tune the sparse Speeded-Up Robust Features (SURF) flow. It not only reduces the influence of illumination changes but also makes the distribution of optical flow more uniform. Based on sparse SURF flow with multilayer bucketing screener, a complete scheme for the ego-motion estimation in monocular vision systems is proposed. Taking the advantage of two-view estimation to obtain the relative scale, we use the ground plane estimation to calculate the absolute scale. Random sample consensus-based outlier rejection schemes are applied to reduce the scale drift. Then experiments are implemented to evaluate our method in real environments. We tested four data sets, and the results show that our method gains smaller average errors than the other two.