Three-dimensional Object Detection Method on Railway Track using Stereo Camera.

Abstract

We are developing a method to detect abnormal events on and along the rail track using cameras and sensors to assist train crew in monitoring the train ahead. In the developed method, the location of the track and the type of object are recognized from images taken by a monocular camera, and the shape of the object is captured by a stereo camera or LiDAR. In this study, we have developed an algorithm to detect three-dimensional objects with a certain height around the rail track by combining the information on the rail track recognized by the monocular camera and the information on the space acquired by the stereo camera. This paper reports the details of the proposed method and the results of performance evaluation for objects of known size.