전방 추돌 경보 시스템을 위한 비전 기반 차량 검출 방법

Abstract

We present a novel vehicle detection method for forward collision warning system using a single camera. For real-time vehicle detection, the proposed method consists of two steps: hypothesis generation where the locations of possible vehicles in an image are hypothesized and hypothesis verification where tests are performed to verify the presence of vehicles in an image. In hypothesis generation step, we extract quickly candidate regions of possible vehicles using a combined feature of edges, intensity, and geometric information. Then, extracted candidate regions are determined whether vehicle or non-vehicle in hypothesis verification step. In hypothesis verification step, we use our novel AdaBoost classifier with clustering to classify vehicles which have various appearances, and also multi-frame information to remove false positive detections.