Sensing the driving environment with smart mobile devices

This paper presents a generic obstacle detection system using single camera image sequence processing, for Android smart mobile devices. The algorithm relies on transforming the images into bird eye views of the scene using Inverse Perspective Mapping, followed by intensity based segmentation. The obstacle area hypotheses are used for updating a particle based occupancy grid. The sensors built in the mobile device are used for obtaining the speed and the yaw rate of the host vehicle, information needed for compensating the motion of the host vehicle in the tracking process. The occupied cells resulted from occupancy grid tracking are grouped as obstacles having position, length and width, and a speed vector. The system's intrinsic parameters are calibrated using the information provided by the Android API, and the extrinsic parameters are calibrated using an easy to use graphical interface. An automatic calibration method for the pitch angle, relying on the cumulated density of horizontal edges on the image rows is also provided. The system was tested on multiple devices, in real urban traffic, and is able to sense most obstacles on the road.