Adaptive Video-Based Algorithm for Accident detection on Highways

For the past few decades, automatic accident detection, especially using video analysis, has become a very important subject. It is important not only for traffic management but also, for Intelligent Transportation Systems (ITS) through its contribution to avoid the
escalation of accidents especially on highways. In this paper a novel vision based road accident detection algorithm on highways and expressways is proposed. This algorithm is based on an adaptive traffic motion flow modeling technique, using Farneback Optical Flow for motions detection and a statistic heuristic method for accident detection. The
algorithm was applied on different collected videos of traffic and accidents on highways and results prove the efficiency and
practicability of the proposed algorithm using only 240 frames for traffic motion modeling. This method avoids the recurring to a large database while adequate and common accidents videos benchmarks do not exist.