This paper describes a proposed new vehicle detector, using closed circuit television (CCTV) cameras, and presents the results of some field tests on it. The detector uses a robust pattern recognition algorithm, called the directional temporal plane transform method (DTT), together with computing hardware specially designed to process real-time traffic flow managements. The DTT performs a transformation from spatio-temporal grey-level image data to two-dimensional data on a direction-time plane. Data about vehicle features are projected onto a directional axis, which is roughly parallel to the moving loci, and the projected data are placed side by side in time order. The process for vehicle for vehicle recognition and measurement is described. Normally, the CCTV camera is placed high on a roadside pole, and aimed along the observed lane; the vehicle detector is fixed to the pole. In the experimental test, traffic flow on a two-lane road was detected and measured by a CCTV camera 7.0m above the road's centre. The experiment was conducted for periods of 20min in daytime, twilight and night in fine weather, and high recognition rates were obtained in all three cases. The DTT can also detect traffic accidents and jammed traffic.
Abstract