Vehicles Recognition Method using Spatio-Temporal Images

Abstract

Traffic flow measurement sensors using image processing are being installed. We have developed Directional-Temporal plane Transform (DTT) which transforms spatio-temporal images to 2D data on a directional-temporal plane as a vehicle recognition method. In the method, however, it is necessary to set the processing area for each lane to cover the moving paths of passing vehicles. Accordingly, the results of the method are affected by vehicles driving in other lanes under some camera conditions. In addition, the position accuracy of the method in the crossing direction is not sufficient, and as a result, it is unable to detect vehicles changing lanes. In this paper, we propose a vehicle recognition method that utilizes DTT and adaptively changes the processing areas by comparing the position of the areas with the position of 3D vehicle models. We also show the experimental results using real images.