Signboard Detection using YOLOv5 with Transfer Learning
Uday Kulkarni, S. M. Meena, Akshay R. Shanbhag, Amruta Ayachit, Prajakta G. Kulkarni, Shivani Halemani, Supreeth S Kadappanavar · 2023
With the ever increasing popularity of autonomous driving technology, effective and precise detection and identification of traffic signs is critical for subsequent decision-making and control actions. The huge elevation in technology has made it more feasible to classify and detect images. The precision and inference time are the key factors for object detection in real-time. The introduction of high-performance computing resources, cutting-edge deep learning, and machine learning algorithms offer real-time solutions to the above issues. After a detailed survey and study that exhibited different methods to address these issue, we have proposed a deep learning model which is based on one of the recent versions of YOLO. In this paper, we have proposed YOLOv5 with transfer learning as an optimum solution to address the mentioned issue. The proposed method has various potential parameters that turn out better than the other models of deep learning including the older versions of YOLO. The dataset is taken from ’roboflow’ platform consisting of 19 different classes of signboards, each of 1272 images. The proposed and implemented model gives overall high performance and detection accuracy. This model gives 96% accuracy and 0.93 mAP. The model can be equipped with ADAS for better navigation and safety.