India ranks among the countries with the worst road safety record. With over 1.5 lakh road fatalities in the country, safety is one of the top priorities of the government in the road sector. "Fault of Driver" was reported to be the major cause of the accidents. Additionally, various traffic characteristics like weak lane discipline, multiple vehicle types, aggressive driving, and manoeuvering through gaps between the vehicles exist on Indian roads. These characteristics make driving behaviour modelling complex and challenging. There is a need to understand the driving behaviour and minimize driving errors to improve road safety. This can be achieved through in-vehicle sensors that assess driver actions and manoeuvers. From the literature, it is noted that only a few naturalistic studies have been conducted on Indian traffic to characterize driving behaviour. However, the sample size of drivers in these studies was small. Also, the literature did not reveal any study in India focusing on risk indicator thresholds and models.
The overall goal of this study is to collect naturalistic driving data on vehicle trajectories and use it to characterize mixed traffic driving behaviour. Towards this goal, the following objectives are framed. The first objective is to plan, collect and collate naturalistic driving data and utilize this data to analyze the kinematic and traffic-related driving indicators like speed, acceleration/deceleration, and travel time across various road types, sections, and traffic volume periods. The second objective includes extracting risk indicators from the kinematic indicators and identifying the critical driving events based on these indicators and their thresholds. The third objective is to classify the drivers into different driving styles using the clustering technique. The final objective models how the risk indicators vary with the various road, driver, and section characteristics. This study's findings can be incorporated into driver assistance technology, which can warn and assist drivers in promoting safety. The proposed models will be helpful in the early identification of risk-prone drivers.