Vehicle Make and Model Recognition using Deep Transfer Learning and Support Vector Machines

Abstract

Vehicle make and model recognition (VMMR) is a vital task in applications like intelligent parking systems, driver assistance systems, and automated toll collection. Accurate implementation of VMMR becomes challenging due to the similar appearance of different vehicle models. In this paper, we proposed a VMMR framework based on deep feature extraction from VGG16 convolutional neural network (CNN) model, followed by the feature reduction and classification. First, deep features were extracted from the image of the vehicle through the FC-6 layer of VGG16 CNN. Next, the Genetic Algorithm was applied to identify best-describing features and reduce feature dimensions. Finally, a reduced feature vector was supplied to support vector machines classifier to distinguish among various make and models of vehicle images using 5-fold cross-validation. The proposed technique obtained an average accuracy of 98.22%, sensitivity of 94.57%, and specificity of 96.86%. A comparative analysis with state-of-the-art methods shows significant improvement of the proposed model.