Dynamic Scene Modeling for Automatic Traffic Data Extraction

Abstract

Motion analysis in a road scene from a sequence of T.V. images is the key problem for automatic data extraction. The task of estimating the motion parameters requires the computer to identify moving vehicles and pedestrians in each frame of the sequence. In this paper, we propose a new technique for dynamic road scene modeling. Moving edges are first detected by combining differencing and differential operations on successive frames of the sequence, a technique that is quite insensitive to the variations of the illumination of the scene. The knowledge of the boundaries of the moving objects is then used to generate minimum bounding octagonal models of vehicles and pedestrians. Thanks to this new model, it is possible to locate simply any moving object in any frame of the sequence of images. The algorithm, implemented on a high‐speed image processor, has been tested using videotapes of real‐world urban street crossings under various natural lighting conditions.