Vision-Based Lane Crossing Point Tracking for Motorcycles

In this paper, we investigate a vision-based approach for online lane change prediction and detection dedicated Powered Two-Wheeled Vehicles. The approach is composed of two steps. First, the road geometry (clothoid model) and the motorcycle position with respect to the road markers are deduced based an inverse perspective mapping algorithm. The relative position is represented by the vehicle lateral displacement and heading estimated by means of an Inertial Measurement Unit and a monocular camera. The second step consists of predicting the Lane Crossing Point which allows to predict the distance and time before the motorcycle crosses the lane. The algorithm is achieved without the use of any steering sensor. To assess the effectiveness of the proposed approach, the estimation and the prediction schemes are validated on the BikeSim framework. To this end, two scenarios are discussed : 1-straight road with non-zero relative heading, and 2-curved road and circular vehicle trajectory.