Development of Real-Time Detection of Philippine Traffic Signs Using YOLOv4-Tiny
Ryan Christopher B. Pinca · 2024
Object detection is critical in computer vision, with real-time applications spanning autonomous systems, surveillance, and augmented reality. This study addressed the challenge of real-time object detection, focusing specifically on traffic sign recognition, which is essential for road safety and intelligent transportation systems. Leveraging the YOLOv4-tiny algorithm, we developed an advanced machine-learning model tailored for this purpose. The study began with data collection from a diverse dataset of over 300,000 images depicting Philippine traffic signs across various real-world scenarios, including daylight, rain, shady weather, and traffic congestion. Our model demonstrated detection confidence levels of 52% to 99% in these varied conditions, achieving an overall mean average precision (mAP) of 87%, precision of 90%, recall of 95%, and F1-score of 93%. YOLOv4-tiny proved effective for real-time traffic sign detection, with response times ranging from 0.03 to 0.1 seconds across different hardware platforms. This study presented a comprehensive approach to real-time traffic sign detection, promising enhanced road safety and efficiency. Recommendations included collaboration with transportation authorities for real-world validation, standardization of traffic signs in the Philippines, and continuous refinement.