Rapid automatic vehicle manufacturer recognition using Random forest

Abstract

This paper studies the applicability of machine learning methods in identifying the individual vehicle attributes based on camera images from the real environment. We focus on a vehicle manufacturer recognition. Classification based on the front vehicle mask makes possible to identify also vehicles without manufacturer's logo. The algorithm has been evaluated on 2988 samples collected directly from cameras in real environment. Random forest algorithm has achieved the best results in classification. Accuracy for classifying the most frequent two manufacturers, Skoda and Volkswagen has been 97.21% and 98.10% respectively. It is also fast enough to use it in real-time, even on low-cost devices like mobile phones or single-board computers like Raspberry Pi. Functional implementation of this method has been successfully deployed in a real-world environment.