Clustering Method of Driving Scenes based on Road Shape by using Road Map Images

Currently, vehicles are equipped with various sensors that can collect a series of driving behavior data while driving. The driving behavior data obtained in this way can be used to analyze the causes of traffic accidents and evaluate driver skills. However, when performing the above analysis, a specific driving scene is defined and driving behavior data for that driving scene is required. On the other hand, the series of driving behavior data obtained during one driving operation is enormous, and it is not easy to pick up desired driving scene data from among them.This research aims to establish a driving scene clustering method for a series of driving behavior data. In this paper, a driving scene is defined as a scene in which road shape and operation are associated, and propose a driving scene clustering method considered road shape. Here, a road map image is acquired from position information obtained from a GPS sensor, and the relationship between the road map image and operation data is clustered based on Coupled-GP-HSMM.

I. Introduction

When analyzing a traffic accident, a specific driving scene to be analyzed is defined [1], [2]. In this case, it is necessary to collect driving behavior data for the defined driving scene. However, manual data collection requires a great deal of effort. On the other hands, it is easy to obtain driver’s operation data, vehicle’s position data, situation data around a vehicle because most of vehicles are equipped with CAN(Controller Area Network) bus module, GPS(Global Positioning System), and in-vehicle camera. Therefore, in this paper, we propose a clustering method for a series of driving behavior data. Our method is clustering of driving scenes such as turning right at an intersection or passing a vehicle parked on the street. The realization of this technology will contribute to streamlining the work of collecting driving scene data necessary for analysis.