Video Traffic Analysis for Abnormal Event Detection. Innovation Gap Research Grant Pre-proposal 2007- 2008. Final Report

摘要

We propose the use of video imaging sensors for the detection and classification of abnormal events to be used primarily for mitigation of traffic congestion. Successful detection of such events will allow for new road guidelines; for rapid deployment of various transportation and safety agencies; for interactive displays that alert drivers to, for example, slow down or speed up, move to a different lane, or, alter their driving behavior, so as to reduce the probability of traffic congestion or the occurrence of more abnormal events. Deciding on additional road guidelines or proper display information, can be accomplished either via computer simulations or experimentation with field implementations. We extended in-house developed algorithms to detect and track vehicles in video sequences and analyze their trajectories. Through analysis we will classify their trajectories independently but also considering vehicle interactions, into abnormal and normal events. A main objective of the analysis is to determine how each type of abnormal event affects subsequent traffic, and serves as a predictor of congestion build up. Towards this task we will identify each type of abnormal event by implementing a supervised classifier, and built models to describe them and their effect.