Automatic incident detection algorithm based on support vector machine

Abstract

Incidents contribute to a significant proportion of delays and costs to the motoring facility. The rapid detection and clearance of incidents is one of the most effective means of reducing the impacts of such non-recurring events. Automatic incident detection (AID) is an important essential component of an Advanced Traffic Management and Information Systems (ATMIS). According to the analysis of the traffic data change under different traffic conditions, velocity and occupancy rate change clearly, while the change of traffic flow is not obvious. This paper proposed automatic incident detection algorithm based on support vector machine (SVM) for freeway. The input of SVM is selected as the upstream velocity, the incident point velocity, the downstream velocity, the upstream occupancy rate, the incident point occupancy rate and the downstream occupancy rate. The output of SVM is traffic state (incident or non-incident). The AID algorithm based on SVM is compared with the artificial neural network. The experimental results confirmed that SVM is a superior pattern classifier for AID. The results suggest that AID algorithm based on SVM has the higher potential for use in an operational automatic incident detection system for freeway.