Vehicle Brands and Types Detection Using Mask R-CNN
Mohammad Wahyudi Nafi'i, Eko Mulyanto Yuniarno, Achmad Affandi · 2019
Detection and classification of vehicles are inseparable parts of Intelligent Transportation Systems (ITS), various kinds of information technology applications are used to be able to detect and classify these vehicles, starting with the use of ultrasonic sensors, laser scanners, induction loops, magnetic sensors, range sensors, pressure sensors and CCTV cameras, but the circulation of vehicles with the same design from different manufacturing companies makes the classification of vehicles to determine the vehicle brands and types difficult to do. In this paper, deep learning framework Mask Regional-Convolutional Neural Network (Mask R-CNN) is used to solve the problem. Experiments have been conducted twice by using a combination of different datasets and detection algorithms. To be able to distinguish cars with similar shapes from different manufacturers, we use vehicle logos as one of the features that distinguish the manufacturer. The best detection and classification results were obtained in dataset training using 60 epoch, 400 step iterations with an accuracy value of 0.91 and mAP (Mean Average Precision) of 0.89.