Deep Learning-Based Vehicle Type Detection and Classification

Abstract

Modern intelligent transportation systems heavily rely on vehicle-type classification technology. Deep learning-based vehicle-type categorization technology has sparked growing concern as image processing, pattern recognition, and deep learning have all advanced. Over the past few years, convolutional neural work, in particular You Only Look Once (YOLO), has shown to have considerable advantages in object detection and image classification. This method speeds up detection because it can predict objects in real time. High accuracy: The YOLO prediction method produces accurate results with low background errors. YOLO also understands generalized object representation. This approach, which is among the best at object detection, outperforms R-CNN approaches by a wide margin. In this paper, YOLOv5 is used to demonstrate vehicle type detection; the YOLOv5m model was chosen since it suits mobile deployments, the model was trained with a dataset of 3000 images, where 500 images were allocated for each class with a variety of vehicles. Hyperparameter tuning was applied to optimize the model for better prediction and accuracy. Experimental results for a batch size of 32 trained for 300 epochs show a precision of 98.2%, recall of 94.9%, mAP@.5 of 97.9%, mAP@.5:.95 of 92.8%, and overall accuracy of 95.3% trained and tested on four vehicle classes.