Abstract
Traffic flow surveillance systems are used to maintain comfortable driving and road safety. In this paper, we propose a traffic flow surveillance method using a spatio-temporal image processing that provides low cost traffic flow classification and can be applied to systems that are equipped with numerous cameras. This method uses the DTT method, which transforms spatio-temporal images into 2-D data (DTT image) on a directional-temporal plane, and creates an edge-directional histogram of the DTT image. Using the histogram as input signals, the traffic flow is classified into five patterns (Normalcy, Traffic congestion, Heavy traffic congestion, Stop, and Low speed) through a neural network. To create teaching data for the neural network, we used the data obtained by shifting the histograms for standard patterns and by making computer graphics images. Experimental results show the effectiveness of the method. In addition, we describe a low cost system, which is equipped with intelligent cameras that have simple image processing functions, using the proposed method.
