Obstacle Detection using Camera and LiDAR for Train Forward Surveillance

For the further improving the railway safety, we are developing a train forward surveillance method using cameras and sensors.In train forward surveillance, it is important to detect obstacles entering the tracks at a distance.Therefore, we investigated the combination of images obtained from cameras and point cloud data obtained from LiDAR as a suitable senser configuration for the railway environment.In our proposed method, the detection area is set by predicting the area of the rail track from the image, and objects such as people and automobiles are recognized using the deep learning.In addition, by combining detection using point cloud data, we can avoid missing objects even in conditions where recognition from images is difficult, such as at nighttime.