Front collision warning based on vehicle detection using CNN

Abstract

Front Collision Warning(FCW) is a critical safety function of Advanced Driver Assistance System(ADAS). Recently, many researches related to FCW systems which use monocular camera image processing have been introduced. In this paper, we propose an FCW system for highway environment based on vehicle detection using Convolutional Neural Network(CNN) as a classifier. Adaptive Region-of-Interest(ROI) is set using lane detection to enhance speed and detection performance of the system. We measure the distance between our vehicle and the detected vehicle in front by calculating the ratio between the lane width of the position of the detected vehicle and our vehicle, respectively. Time-to-Collision(TTC) is used as a collision warning index. For FHD(1920×1080) black-box camera images taken in highway environment, the detection rate of the proposed CNN is 99.1%, and the execution time of the system is 19.8ms per frame.