Traffic Incident Detection and Analysis System (TIDAS)

active

Existing highway incident detection technologies fail to use available footage from traffic cameras installed on the highways. On the other hand, monitoring and analyzing the overwhelming quantity of camera data without assistive automated methods is intractable. Using Artificial Intelligence (AI), models can be trained to enhance images and provide robust detection and classification of traffic incidents, resulting in more cost-effective deployment of incident response resources. This research focuses on solving challenges including: 1) the lack of a robust automatic incident detection system capable of emphasizing key events with minimal false alarms, 2) overcoming the problems inherent in current learning algorithms, which significantly degrade in performance under adverse weather conditions, 3) the lack of availability of a dataset with diverse footage of highway incidents to foster the development and validation of AI algorithms. Here, a novel framework using AI and image processing algorithms for exploiting the potential of currently installed highway camera infrastructures for highway incident detection is presented, including detecting wrong-way driving, traffic congestion, crashes, and bicyclists or pedestrians in tunnels.