A spatiotemporal learning approach to individualized driving risk assessment in a Vehicle-to-Everything (V2X) environment

Keywords: V2X, Risk Assessment, Spatiotemporal
Abstract Type: Paper Abstract

Abstract

The recent Vehicle-to-Everything (V2X)-enabled infrastructure delivers massive Basic Safety Message (BSM) data to describe states of sensor-equipped vehicles in a real time fashion. This paper presents a novel machine learning based risk assessment method that leverages BSM from a V2X infrastructure to measure individualized driving risk, which is defined as a function of probability and frequency of Safety Critical Events (SCEs) in the immediate driving environment of a driver at a time point. Focusing on the overall safety situation, this assessment method generates full depictions of driving environments with vehicle states provided by BSM data and road and weather conditions from public repositories. A combinatorial feature learning and clustering model is introduced to extract features from the high dimensional resignation of driving environments and establish the connections between features of driving environments and occurrence of SCEs to quantify individualized driving risk levels. This method further the efforts of providing fine level and cost-effective risk assessment solutions to individual drivers. To demonstrate the usage of the assessment method, a case study was conducted using the pilot V2X infrastructure deployed in Tampa, Florida. Results show that the proposed approach can yield high accuracy in determining different driving risk levels.