Two-Step Vehicle Classification System for Traffic Monitoring in the Philippines
Rodrigo N. Celso, Zachary B. Ting, Dale Joshua R. Del Carmen, Rhandley D. Cajote · 2018
Traffic congestion is a major problem in the Philippines. A vehicle classification system, for a proposed Philippine metropolitan traffic monitoring system, would be useful for various applications such as: road monitoring, law enforcement, and emergency vehicle prioritization. This paper implements a two-step vehicle classification system that classifies vehicles into three vehicle sizes and seven vehicle classes. Vehicle images are first extracted using the ViBe background subtraction algorithm on a Philippines metropolitan traffic sequence. The dataset used is comprised of properly segmented vehicle images. Geometry-based and texture-based features were extracted from the isolated image of each vehicle. Feature selection is used to determine the best features to be used in each classifier. Various machine learning models (kNN, SVM, MLPN) are used to implement the two-step vehicle classification system. The implemented vehicle size classifier uses a kNN model with an F-measure of 97.41%. Next, the medium vehicle classifier uses an SVM model with an F-measure of 90.68%. Finally, the large vehicle classifier uses an MLPN model with an F-measure of 97.97%. The resulting two-step vehicle classification system has an overall F-measure of 91.41%. This designed two-step classification system assumes that the input vehicle is properly segmented by background subtraction.