An Innovation-Based Approach to Timely and Robust Automatic Highway Incident Detection

Abstract

Timely and accurate detection of traffic incidents is of crucial importance for highway management and warning systems. These tasks are normally carried out by algorithms of traffic state estimation (SE) combined with automatic incident detection (AID) on the basis of local measurements (e.g. inductive loop data, radar data, etc.). It is evident that faults and inaccuracies in the process of measuring traffic data affect the quality of SE and AID significantly. Especially faults in measuring the traffic volume occur frequently. They cause false alarms or they hinder these systems to detect dangerous incidents. In this article we apply methods of technical fault diagnosis to design an AID. Therein, a vector-based innovation of an extended Kalman filter (EKF) is used for distinguishing traffic incidents from flawed data and from other disturbances on traffic flow. Also it is used for timely detection of traffic incidents. For reasons of a robust application in real-world scenarios with flawed data, some modifications are discussed and their positive effects on the innovation-based approach are presented.