Deep Learning based Traffic Direction Sign Detection and Determining Driving Style

Abstract

Intelligent automobiles and advanced driver assistance systems (ADAS) are some of the major technological developments that affect human daily life. Today, many studies are being generated to develop state of the art transportation systems. The general objective in these studies is to cope with negative effects of traffic. In this work, our aim is to contribute to the development of ADAS by determining driver behavior and traffic direction sign detection. The data employed are acquired by smartphone sensors, which are accelerometer, gyroscope, GPS, and camera, while the subject car moves between two specific points. The proposed method consists of two simultaneously running algorithms. The first one determines driver maneuvers, and the second one is the deep learning based algorithm that detects traffic direction sign using Convolution Neural Network (CNN). Here, the results of these two simultaneously running algorithms are assessed, and driving type is determined. GPS data is used for synchronization. Consequently, it is determined whether riding style is safe or aggressive, involving in traffic direction sign detection.