Development of a Rain Sensor Using the YOLO Model

This study deals with developing a rainfall detection system that uses the YOLO (You Only Look Once) object detection model to recognize raindrops falling on vehicle windshields and automatically control the wipers. Data were collected by simulating rainy conditions, and used for training the model. The detection performance was evaluated for various conditions, including confidence threshold, GPU performance, camera resolution, and distance variations. The results showed that the optimal confidence threshold (0.28) and distance range (6.5 to 7.5 cm) were identified. It was also found that detection performance and stability significantly improved when using a high-resolution camera (1080p) and a high-performance GPU (NVIDIA RTX 3060). The YOLOv5-based detection system demonstrated high accuracy and automation levels, overcoming the limitations of traditional sensor-based methods. Further research is needed to optimize performance under diverse environmental conditions and verify its potential for real-time applications.