Abstract

GPS signals, while typically noisy, can be very usefully exploited
to help understand attributes of an underlying road
network. In this paper we present the results of our GPS trace
classification, a method that analyzes GPS signals collected
around an intersection and categorizes its traffic-control system,
such as stop-signs or a traffic-light. We represent an
intersection as a feature vector of speed variations obtained
from GPS traces and use these feature vectors to learn the
models of different systems of traffic-control. In combination
with a geographic distribution of intersections of different
traffic-control systems, we assign a previously unseen intersection
as the most probable traffic-control system. Experiments
show a promising result of our classification task
applied to real-world GPS trace data.