University of California Transportation Center

Knowledge of vehicle classes is especially useful for monitoring commercial vehicles (CVs). Accurate CV class information will enhance truck traffic surveillance and fleet management, such as in port areas by providing information for environmental impact investigations. From an implementation perspective, it is recognized that there are often significant advantages to use the existing inductive loop infrastructure. However, inductive loops are not always the most practical surveillance technology considering the required implementation effort and cost. In this regard, this study explored the potential of adopting a new vehicle signature detection technology - wireless magnetic sensors - for CV classification. The vehicle signature data used for the development of the wireless sensor based models was collected from the University of California, Irvine (UCI) Commercial Vehicle Study Test-bed in San Onofre, California. Vehicle signatures from round inductive loop sensors were also collected for refining an existing round loop based model and for comparison purposes. Significant dropped data was observed in the wireless sensor signatures, which required the implementation of a dual sensor data recovery procedure to reconstruct the signatures, which would otherwise have been unusable. The results indicate that the single wireless sensor vehicle classification model, which is based on multi-layer perceptron neural network, successfully distinguished single-unit and multi-unit trucks with 93.5% accuracy. The double wireless sensor vehicle classification model, which adopted a K-means clustering and discriminant function, achieved 73.6% accuracy, while the round loop based model produced even better performance (85%) in testing, both according to the FHWA scheme F with 13 classes.