Link speed estimation using GPS data: an empirical investigation of some issues

Abstract

Probe vehicles equipped with tracking devices such as global positioning system receivers (GPS) can be utilized for real-time link speed estimation. In this empirical research, the impact of the data collection resolution, also known as the polling interval, on the network coverage and link speed estimation accuracy was explored. Furthermore, a comparison was made between different methods that currently exist for average link speed estimation using GPS data. The study made use of a 1 s resolution GPS dataset that covered 100 trips in Vancouver, BC. The dataset was sub-sampled 36 times to simulate cases of 5–180 s sampling intervals. An existing map-matching algorithm was used to match the GPS points to the correct travel links for the 36 datasets. Consequently, average link speed was calculated for each link in the dataset using the time stamp difference method and the average instantaneous speed method. A slight variation of the average instantaneous speed method was also tested where instantaneous speeds were computed from position information only. An improvement was further applied to the latter method by using a path inference technique to compensate for the lack of GPS points on some links. The speed estimation methods were compared at different polling intervals and the results were discussed. In general, it was shown that the average instantaneous speed method provides the highest estimation accuracy while the path inference method provides the highest coverage compared to all other methods.