Map Data Integration Technique with Large-Scale Fleet Telematics Data As Road Safety Surrogate Measures in The New York Metropolitan Area

Conventional road safety models rely on historical crash data. Locations with high crash injury statistics are given primary interventions. However, crash data are subject to errors, under-reportings, inaccuracy, and requires years to get updated, as crash events are infrequent and partly random(Gettman, Pu, Sayed and Shelby, 2008), as well as road conditions might change. With the advances in connected vehicle technologies, vehicles can be used as mobile sensors that collects driver behavior information. And if found correlated with the crash data, the driver behavior indices can act as safety surrogate measures. This dissertation focuses first on presenting an algorithm for mapping a vehicle sensing big dataset to a digital road network, in a metropolitan city, using the accompanied GPS trajectories. This is a challenging task since the data collected from the on-board-diagnostic port of the vehicle is sampled at a low ping rate, adding to that the excess of GPS noise in urban canyons, which makes the map matching problem even harder. Next, a spatial correlation study is presented. It investigates the spatial relationship between the driver behavior indices (i.e. speed parameters, hard braking and hard acceleration) and crashes (crash frequencies and crash rates, normalized with traffic volume). Highways and non-highway roads are bucketed separately. The other focus of this dissertation is developing an injury-prediction network screening model, that provide safety ranking of road intersections. Novel explanatory variables are derived from the telematics data, such as intersection traffic maneuvers and traffic conflicts. The non-linearity between the explanatory variables as well as the spatial dependency between road intersection is also tested.