고속 자율 주행 차선유지시스템을 위한 DNN 기반 레이더를 이용한 차선 추정

Abstract

In this paper, we propose a novel Deep Neural Network(DNN)-based lane estimation method with radar for lanekeeping system(LKS) without vision sensors. The vision sensor is widely utilized for detecting the road lane. However, it is well known that the vision sensor has weaknesses about environmental effects such as weather, contaminated lane, and so on. Thus, we exploit the radar sensor, which is robust against environmental effects, to estimate the road lane model for LKS. Scaled Conjugate Gradient(SCG) method is used for optimization of the neural network. We had a comparative study between the proposed system and the camera system which is conventionally used for LKS. The proposed system is expected to improve the performance of LKS by sensor fusion.