Modelling the accident occurrences per unit length of the road requires segmentation of road stretch. In the case of hilly terrains, homogeneity-based approach results in extremely small sections thus creating a biased accident data. This result is due to the high variations in geometry along the length of the road stretch. Constant length sections do not consider the effect of road's characteristics, hence may lead to undesirable results. It is hypothesized that the speed profile of a vehicle moving in free-flow conditions is governed by the road geometry as well as the road side environment. In this paper, a new methodology based on speed profile of a test vehicle (passenger car) has been proposed for segmenting a road stretch passing through hilly terrain. Comparative analysis of the results obtained from the constant length segmentation approach and the proposed approach shows that the later approach is producing statistically better results. Relatively more parameters were found to be statistically significant in the model based on speed-profile-based segmentation. Grade length, which could be one of the major parameters in hilly terrains, was found to be statistically significant only in the model based on speed-profile-based segmentation.
Keywords: Poisson model; accident frequency; free-flow condition; segmentation; speed variation.