Video incident detection tests in freeway tunnels

European regulations for the management of roadway tunnels became much more stringent after the 1999 Mount Blanc inferno. Several Video image processing (VIP) devices for traffic surveillance have automated incident detection (AID) capabilities. Autoscope, Citilog, and Traficon were invited to make installations for testing their VIPs with eight preexisting cameras in Attica Tollway tunnels. The suppliers of the competing devices were responsible for setting and calibrating the devices for best results. No literature was found in which VIPs developed in 2000 or later were evaluated for incident detection. A 3,013-incident database was used to derive the detection rate-and false-alarm rate for each device for the whole database, separately for each closed-circuit television camera, and by type of incident. Changes in detection performance after each manufacturer's intervention were evaluated. T-tests were conducted on the significance of effects of low, medium, and high volume in weekday and weekend traffic, and natural or artificial lighting. A separate analysis was conducted for incidents that were artificially generated by Attica Tollway engineers and staff at off-peak times. The results show promise, but the evaluated performance was poor because of three reasons:, low maturity of the technology, complex algorithms due to the provision of extensive functionality, and suboptimal camera location and height for image processing, but VIP devices must adapt to tunnel limitations.