Convolutional Neural Network를 이용한 주차 차량 자세 추정

Abstract

This paper proposes a technique based on Convolutional Neural Network(CNN) using rear view camera images to estimate the vehicle pose for the target point when parking. Identifying the vehicle’s pose for the target point in autonomous vehicles is a key technology in the automatic parking system. Generally, existing algorithm is used to recognize parking spaces with rear view camera images or rear view camera images and ultrasonic sensors to identify the vehicle’s poses for the target point. In this paper, we present feasibility of the network learned by CNN as AVM images to find out the pose of the vehicle simply. We show the performance that was observed when confirming the root mean squared error of the actual values and the predicted values.