Optimising automatic incident detection (AID) based on image processing to ensure road tunnel safety: combining performance and robustness

An Automatic Incident Detection (AID) system is an essential tool for detecting events leading up to incidents in road tunnel. Operator expectations are therefore extremely high, as this supporting tool can be used to save time is situations where reacting quickly is crucial to protecting users. To ensure that it works properly, thought must first be put into the system ahead of time to determine the events to detect. Furthermore, the system must be carefully calibrated to achieve the best compromise between a minimum non detection level, the minimum number of false alarms and a suitable detection delay time. This paper serves to present how to harden a tunnel AID system so as to achieve optimum performance. This means covering the choice of incidents to detect, where to locate the cameras, the calibration phase and what maintenance is recommended. Answering these questions leads the way to an AID system offering the best possible performance.