A Novel Distance Estimation Method Leading a Forward Collision Avoidance Assist System for Vehicles on Highways

Abstract

This paper proposes a novel distance estimation method to build a forward collision avoidance assist system (FCAAS) containing techniques of lane marking detection, vehicle tracking, and distance estimation. First, a lane marking detection technique uses a RANSAC algorithm to extract lines of lane markings, which were previously collected from an inverse perspective mapping image filtered by steerable filters. A Kalman filter then tracks the extracted lines accurately and efficiently. Second, a vehicle tracking technique implements a multiple-vehicle tracking method using a particle filter, which tracks the vehicles detected by an AdaBoost classifier. An improved particle filter is implemented to predict the next movement of a vehicle and spread the particles near the predicted location of the vehicle instead of originally spreading the particles around the current location of the vehicle. Finally, an innovative distance estimation method is derived to estimate the distance between the ego vehicle and the front vehicle. The distance estimation method is verified by setting several standard points in the image, whose locations can be measured according to the regulation of lane markings. As a result, verification of the distance estimation method demonstrates a robust feasibility in reality. The FCAAS shows its potential in particular scenes through many experimental sequences acquired from highways in the real world. In addition, the FCAAS fits the demand of a real-time speed system with a speed of 22 frames/s.