카메라를 이용한 Deep Convolutional Neural Network 기반 자율주행차량의 다중주기 횡방향 모션 제어

Abstract

In this paper, a deep convolutional neural network that predicts the road coefficients by training the forward driving area images is proposed. In order to acquire the input image used for training, a low-cost dash camera was mounted to the windshield. Through the proposed algorithm, it is verified that the appropriate control input is generated even in the situation where the conventional lane keeping system fails. The verification was conducted through computational simulation with multi-rate state estimator.