Vehicle Make and Model Recognition Using Structural and Pattern Based Feature Descriptors

Vehicle Make and Model Recognition (VMMR) is gaining popularity nowadays due to prevailing security situation, as it plays an important role in surveillance and forensics. The main issue one must tackle is the fact, that proposed VMMR algorithm must be able to distinguish between different makes and models, while classifying different variants of a model as one. In addition, the algorithm must consume very less time. In this paper, VMMR problem is solved using four different feature descriptors including Local Tri-Directional Patterns, Histogram of oriented gradient (HOG), Wavelet and Curvelet transformation. In addition, we have deployed two different classifiers in our paper to assess their recognition rate while using same feature vectors. These classifiers are Linear Support Vector Machine and K-nearest neighbor. SVM classifier gave an accuracy of 95.67% while the KNN gave an accuracy of 95.89% on BMMRv1 dataset. Achieved accuracy was compared with various contemporary VMMR solutions.