Improved Design for Vision-Based Incident Detection in Transportation Systems Using Real-Time View Transformations

Abstract

Advances in machine vision techniques have led to algorithms and integrated systems that can be applied in transportation engineering to improve surveillance and control. Despite these advances, certain problems in the effective integration of machine-vision based systems at complex intersections and complex freeway sections still remain. These are related to increasing system performance in the identification, analysis, and detection of the traffic state in real time. This work examines the feasibility of providing transformed visual input to existing machine-vision based systems, in order to gain increased efficiency and cost effectiveness of integrated transportation systems. Two transformations are developed,