Railway running smoothness monitoring based on visual motion parallax

摘要

Railway running smoothness is an important factor for both comfortableness and safety. Though acceleration sensor based monitoring methods have been studied for years. This paper proposed a novel running smoothness monitoring method by utilizing the in-vehicle surveillance camera as the motion-sampler and devised an algorithm to perceive posture variations. Since cameras are common in modern railway transportation, the proposed method has good potentials of broadly applications. The proposed method is composed by following steps. Firstly a bunch of key-points are extracted from consequent images and matched in pair to learn sparse pixel flows. Then the fundamental matrix, focus of expansion and vanishing point are robustly calculated from the pixel flows. Finally the tangential speed perpendicular to the attitude direction and rotational speed are estimated from the image flow, focus of expansion and vanishing point. Promising experimental results demonstrate validations of the proposed method with respect to various situations.