Assessing Crash Risks on Curves

Abstract

In Queensland, curve related crashes contributed to 63.44% of fatalities, and 25.17% required hospitalisation. In addition, 51.1% of run-off-road crashes occurred on obscured or open-view road curves (Queensland Transport, 2006). This paper presents a conceptual framework for an in-vehicle system, which assesses crash risk when a driver is manoeuvring on a curve. Our approach consists of using Intelligent Transport Systems (ITS) to collect information about the driving context. The driving context corresponds to information about the environment, driver, and vehicle gathered from sensor technology. Sensors are useful to detect drivers’ high-risk situations such as curves, fogs, drivers’ fatigue or slippery roads. However, sensors can be unreliable, and therefore the information gathered from them can be incomplete or inaccurate. In order to improve the accuracy, a system is built to perform information fusion from past and current driving information. The integrated information is analysed using ubiquitous data mining techniques and the results are later used in a Coupled Hidden Markov Model (CHMM), to learn and classify the information into different risk categories. CHMM is used to predict the probability of crash on curves. Based on the risk assessment, our system provides appropriate intervention to the driver. This approach could allow the driver to have sufficient time to react promptly. Hence, this could potentially promote safe driving and decrease curve related injuries and fatalities.