Traffic Incident Detection Using Probabilistic Topic Model.
Akira Kinoshita, Atsuhiro Takasu, Jun Adachi · 2014
Tra c congestion is quite common in urban settings, and is not always caused by tra c incidents. In this paper, we propose a simple method for detecting tra c incidents by using probe-car data to compare usual and current tra c states, thereby distinguishing incidents from spontaneous congestion. First, we introduce a tra c state model based on a probabilistic topic model to describe tra c states for a variety of roads, deriving formulas for estimating the model parameters from observed data using an expectation–maximization algorithm. Next, we propose an incident detection method based on our model, which issues an alert when a car’s behavior is su ciently di↵erent from usual. We conducted an experiment with data collected on the Shuto Expressway in Tokyo over the 2011 calendar year. The results showed that our method discriminates successfully between anomalous car trajectories and the more usual, slowly moving tra c. However, our method does sometimes classify abnormally fast-moving cars as tra c incidents.