Multi-type Vehicle Detection and Classification Using YOLOV5

In the modern era marked by technological advancement, vehicle recognition using computer vision has become increasingly important. The ability to accurately identify and classify vehicles has far-reaching implications, from efficient traffic management to better road safety. Some research on vehicle classification have been performed, but for multi-type vehicles still needs to be carried out due to the limited efforts on it, especially in vehicles variation, not only on specific vehicle such as cars, but also for two-wheeled and more. The problems of automated vehicle recognition are the variety of vehicle types on the roads; the ability to recognize many vehicles in real-time; as well as the accuracy of classifying vehicles where many obstacles and occlusions appears in the real-world. Hence, this research develops a web-based application that can recognize multi-type vehicles in real-time. The YOLOv5 method is used because of its advantages in fast recognition time and high accuracy, especially for surveillance data. Testing on dataset of multi-type vehicles including cars, trucks, motorcycles, and bicycles gave results with the accuracy rate of 93.23% and F1 score of 0.93. This research contributes to produce a web-based application for fast detecting and classifying multiple types of vehicles with the superior accuracy rate.