Recognition of vehicles silhouette using combination of classifiers

This article investigates the combination of neural networks in committees for pattern recognition. For such, three neural networks were trained to recognize vehicles using a dataset composed by 946 vehicles silhouette, each silhouette formed by 18 characteristics. After the networks have been trained the results achieved by the individual neural networks were combined using parallel combination. Two type of decisions were used: hierarchical decision and majority decision. The results obtained by each combination were compared with those achieved by the individual networks. The experiments were performed following the Proben1 methodology proposed by Prechelt (1994)