Vehicle Detection and Tracking

Abstract

This paper presents a monocular lane-vehicle detection and tracking system comprising of (i) lane boundary detection, (ii) lane region tracking, and (iii) vehicle detection with a proposed vertical asymmetry measurement. First, a traffic scene image is divided into possible road region. Lane boundaries are then extracted from the region using lane markings detection. These detected boundaries are tracked in consecutive video frames using a linear-parabolic tracking model. Therefore, an approximate lane region is constructed with the estimated model parameters. By integrating the knowledge of lane region with vehicle detection, vehicle scanning region is restricted to the road area so as to detect the shadow underneath a vehicle continuously with less interference from the road environment and non-vehicle structures. A self-adjusting bounding box is used to extract likely vehicle region for further verification. Besides horizontal symmetry detection, a vertical asymmetry measurement is proposed to validate the extracted region and to obtain the center of frontal vehicle. Preliminary simulation results revealed good performance of lane-vehicle detection and tracking system.