A Study on the Application of Contour Features to Improve the Accuracy of Autonomous Driving Recognition Based on Deep Learning

Abstract

With the advent of the era of the Fourth Industrial Revolution, the field in which computers can replace humans is gradually expanding. In particular, parts related to artificial intelligence are in the spotlight, and among them, interest in autonomous driving is increasing. Object recognition has become one of the indispensable technologies in autonomous driving, and just as humans recognize traffic lights and traffic signs, many people are making efforts to make computers recognize these types of objects as accurately as possible through cameras. Therefore, in this paper, a study is conducted on whether the recognition rate of the object will be improved by learning the original traffic sign dataset and the contour extracted from the image together. The Inception model and the ResNet model were used to train the Chinese traffic sign dataset. Furthermore, mask datasets and land use datasets are additionally checked to confirm the consistency of the study.