AUTOMATIC DATA EXTRACTION OF VEHICLE TRAJECTORY BY DIGITAL IMAGE PROCESSING FOR ANALYZING BEHAVIOUR - EXPERIMENTAL RESULTS
Alfredo Garcia, María Elena Díaz · 2000
The purpose of this work is to report the first results of a system that is being developed and tested as a method for tracking motor vehicle trajectories in curves from a sequence of digital images. The data allow one to obtain in an automatic way the spatial and temporal evolution of vehicles, and therefore to analyze and model the vehicle behaviour in curves. Other critical points or sections, such as acceleration and deceleration lanes, intersections and weaving links, could be studied with this technique. Vehicle detection must be as accurate as possible, but not necessarily fast, since real time processing is neither considered nor necessary. The method is based on image processing techniques. A set of relevant points that identify the vehicle is detected. These points are defined as relevant extremes of the object corresponding to a given prestored simple car model. Model fitting permits the accurate position determination of a representative point (geometrical center of the car). A post-matching of relevant points between consecutive pairs of images is done to estimate the movement direction at each location and instant of each vehicle. Classical methods are briefly reviewed and compared with the proposed one. Experimental results of vehicle tracking in traffic images on real road curves are reported. The data are compared in terms of lateral placement, path curvature, speed and side friction demand with the expected ones from the current design criterion and standards. Safety margins can be determined.