Image processing has been applied to traffic analysis in recent years, with different goals. In this paper, an algorithm is carried out for tracking vehicle and extracting vehicular speed information, given a sequence of real-time traffic images.This algorithm extracts moving vehicles using median filter, frame differencing and image Boolean calculation, resumes part of losing pixels using mathematical morphology, calculates the centroids of vehicles by obtaining the minimum exterior rectangle of the moving edges, processes the resulting edge information to obtain quantitative geometric measurements of vehicles. This differs from some other approaches because this algorithm uses simple geometric relations obtained directly from the image instead of using reference objects to perform camera calibrations. The algorithm allows the recovery of the physical descriptions of traffic scenes without explicit camera calibration, extracts scaling signatures using a car length distribution and computes the speed distribution on the basis of the geometric relationships in the image. The algorithm in this paper realizes vehicular average speed statistic in a certain period of time rather than extract the exact speed of every vehicle.In this paper, extensive experiments using images from active freeway cameras are reported. The results presented here demonstrate the validity of this algorithm which requires neither direct camera control nor placement of a calibration object in the environment. This paper further argues that it is straightforward to extend this algorithm to other related traffic applications.
|