Contents I. Introduction
Over many years many practical application of vision for monitoring traffic[5], [7] have been developed, typical motorway surveillance systems generate image from static cameras looking at highly predictable traffic flow. A major objective is to develop automatic means to detect, determine the speed, and classify vehicles as they pass a predetermined point. Most approaches to vehicle tracking and recognition take a strictly 2D approach in which some image attribute, such as the convex hull of region itself, is tracked. The techniques copy poorly with multiple overlapping vehicles, or with shadows and rain, since the image attributes used fail to distinguish a vehicle's structure from other signals. This problem can be greatly reduced by using 3D model-based methods[1],[2] [3] [6].
