A neural network approach to vehicle classification with double induction loops

The paper describes the development of a neural network algorithm for vehicle classification with loop detectors. The basic input for the classification is the loop analog signal which has a form typical to different vehicle classes. Seven classes of vehicles are used, namely car, van or pickup, car/van/pickup with trailer, truck, bus, truck with semitrailer and truck with full trailer. The analog loop signals are first preprocessed to form feature vectors of twenty elements. A 12*12 feature map is trained by self-organised learning with a sample of about 1400 field observations. The resulting weight factors are corrected by vector quantisation. Finally, the vehicle length is included as a limiting factor. The results of a field test indicate that neural networks can be successively used in the vehicle classification. Several possibilities for further development of the method are also available.