Time to collision estimation for vehicles coming from behind using in-vehicle camera

Abstract

In recent years the number of Advanced Driver-Assistance Systems (ADAS) implemented in vehicles has been rising significantly. In this paper an algorithm for ADAS that estimates the time to potential collision (TTC) with the vehicle coming from behind, by processing only the video frames captured by the camera located in the outside rear-view mirror, is proposed. After preprocessing of the input video frame, vehicle detection is performed, followed by vehicle tracking. When the vehicle is successfully detected and tracked, it is possible to estimate the distance between the oncoming vehicle and the camera, as well as the oncoming vehicle speed. Using the calculated distance and the speed of the oncoming vehicle, the proposed algorithm outputs the final information about the TTC, i.e. time remaining for safe change of the driving lane. The algorithm has been tested on different real video sequences and it achieves high performance in terms of estimated TTC accuracy, while processing 25 Full HD frames per second.