Many models have been developed to predict collision frequency and evaluate safety performance on horizontal curves. The approach used in data collection or some assumptions made in the analysis methodology might lead to inaccurate results. For example, manual data collection, equipment limitations, and field experiments involving monitoring driving behavior for a specific region for a short-term are potential sources of errors in data collection. This paper aims at overcoming some of these issues in developing models to evaluate safety performance of horizontal curves and predict the curve collision frequency. The developed models relate expected collision frequency on horizontal curves to the speed reduction from the approach tangent to the curve, which is commonly used as a major geometric design consistency measure. The methodology to achieve this objective included three tasks; data collection, evaluating and modeling the viable speed reduction parameters, and developing safety performance models to estimate collision frequency on horizontal curves. Individual drivers’ trips on 49 horizontal curves on rural two-lane highways in rolling and mountainous terrains in Washington State were extracted from the Naturalistic Driving Study (NDS) database. Models were developed to relate different speed reduction parameters to curve characteristics. These models were then applied to 1430 horizontal curves in Washington State to estimate the speed reduction parameters and relate them to collision frequency. Several safety performance models were developed which show that speed reduction, as a design consistency measure, is directly related to collision frequency on horizontal curves. Furthermore, the speed reduction parameters are more significant variables in predicting collision frequency than all curve geometric parameters.

, , , , , ,
Accident Analysis and Prevention
Department of Civil and Environmental Engineering

Dhahir, B. (Bashar), & Hassan, Y. (2019). Using horizontal curve speed reduction extracted from the naturalistic driving study to predict curve collision frequency. Accident Analysis and Prevention, 123, 190–199. doi:10.1016/j.aap.2018.11.020