Urban Traffic Anomaly Detection

Abstract

The accessibility of large-scale Spatio-Temporal GPS data provides us information for analyzing the traffic system using advanced mining and machine learning techniques. Nowadays, a large number of taxicabs in major cities are equipped with a GPS device so that they can be used to monitor the behavior of traffic on the roads. Here, we are using this GPS data from taxis to monitor the emergence of unexpected behavior in an area. In this work, we aim in efficiently finding traffic anomalies that may occur on the roads i.e. where we want to find the location and time where an accident, road jam, crowd or anything that might hamper the smooth flow of traffic has occurred. We discover these outliers in this GPS dataset using some of the machine learning techniques. Firstly we have used a similarity based method to detect the outliers. We compute a distance based similarity measure to identify the road segments with abnormal properties using some identifiers. Then we have used a density based clustering method, Local outlier factor (LOF) on the same problem, which turned out in efficiently finding the outliers accurately. The novel contribution in our proposed method includes the introduction of time interval and the use of area segments instead of road segments. We validated our approach on Geolife trajectories dataset published by Microsoft research. For the ground truths, we have referred various news sources. We evaluate our proposed method using parameters like precision, recall and F-score.