Highway Vehicle Classification by Probabilistic Neural Networks

The Federal Highway Administration (FHWA) Office of Highway Planning requires states to furnish vehicle classification data as part of the Highway Performance Monitoring Systems (HPMS). To comply with this requirement, most states use the “F-Scheme” to classify vehicles. This scheme classifies vehicles in 13 classes depending on a number of factors, primarily the number of axles and the axle spacings on each vehicle. Classification of highway vehicles using the “F-Scheme” can be automated by properly using visual information of the number of axles and axle spacing; however, this process is hindered by the absence of a suitable logic to be used in the digital computer. Many computer software vendors rely on sharply defined decision trees that are based on the vehicle number of axles and axle spacing, which often results in misclassifying some vehicles. This paper proposes a classification approach that is based on Probabilistic Neural Networks. The paper explains the design of the neural network for this purpose and how to condition the training data. Field results have shown that the proposed network is effective and can classify the majority of the vehicles as defined in the “Scheme F” guidelines and it outperforms the existing decision tree systems.