Recognizing Vehicle Types Using Combined Features from Curvelet Transform and Pyramid Histogram of Oriented Gradients

Aiming to provide information of vehicle types for traffic control and management, an efficient Combined Feature (CF) extraction approach from Pyramid Histogram of Oriented Gradients (PHOG) and Curvelet Transform (CT) is proposed for description of vehicle front images, and Random Subspace Ensemble (RSE) of Linear Perception (LP) classifiers as the base classifier is then exploited for classification of thirteen classes of vehicles, i.e., Audi, Wulin, Chery, Chevrolet, Cityroen, Ford, Changan, Hyudai, Mzada, Nissan, Peugot, Buick, and Toyotas. With features extracted by CF and RSE of LP classifiers on SEU vehicle front image dataset, the holdout and cross-validation experiments are created. Results of holdout and cross-validation experiments show that CF by RSE of LP classifiers outperforms other two classifiers, i. e., PHOG features by LP classifier, CT features by LP classifier. With CF and RSE of LP classifiers, the average classification accuracies of vehicles types are over 90% in both of holdout and cross-validation experiments, which shows the effectiveness of the proposed feature extraction approach and RSE of LP classifiers in developing recognition system of vehicle types. From the confusion matrices of holdout and cross-validation experiments, it is shown that the vehicle class of Mzada has the most recognition accuracy of thirteen vehicles classes, but the vehicle class of Chery is the most difficult to classify of thirteen vehicles classes.