Incidents detection through mobile sensing

Currently, most automatic traffic incident detection systems rely heavily on data collected from fixed sensors such as inductive loop detectors and surveillance cameras. However, fixed sensors are difficult to install and maintain. Moreover, it is quite costly to deploy a large number of fixed sensors to cover a broad area. It is potentially more efficient to use vehicles as mobile sensors to collect traffic data. In this paper, we introduce a new traffic incident detection algorithm which takes advantage of mobile sensors. This algorithm assesses the likelihood of an incident based on the speed data collected by the mobile sensors. Its performance is evaluated thoroughly, in terms of detection rate, false-alarm ratio, and mean time-to-detection, at various combinations of penetration rates and data sampling periods. The results show that the proposed detection algorithm is potentially promising. For a practical scenario with 10% penetration rate and 30-second sampling period, the algorithm is able to achieve around 80% detection rate.