Risk map generation system for intelligent vehicles on community roads via data-driven approach

Abstract

The aim of this study is to develop a risk map generation system on community roads via a data-driven approach. Because there are no roadside cooperative sensing systems on community roads, intelligent motion controls need the prediction of surrounding traffic participants based on a digital map. In addition, to improve the driver acceptance of the system, adaptive motion control considering the risk level at various intersections is desirable. In general, existing risk map preparation methods need too many human resources to prepare the map data for nationwide community roads. To solve this problem, we design a risk map generation system that can be systemized as much as possible for future automated process. First, our proposed system collects the driving data via non-specialized drivers using relatively usual vehicles at various intersections. Next, the system classifies the driving behavior via feature values with regard to the deceleration operation while approaching non-signalized intersections. Based on the driving data of relatively careful drivers, the system extracts the positions of pseudo intersections that are not registered in the digital map. In addition, the system estimates the risk level of intersections based on the driving data of relatively interactive drivers. To confirm the feasibility of the proposed system, we collect the driving data of 45 elderly drivers. According to the evaluation, the system adequately detects the pseudo intersections on actual roads. The estimation results of the risk level show the substantial agreement with the risk evaluations of driving school instructors regarding basic intersections.

Keywords

• automobile
• digital map
• driving behavior
• data driven analysis
• map deepening