Implementasi YOLOV10 Dalam Pendeteksian Rambu Lalu Lintas Dengan Integrasi Teknologi Text-to-Speech

Abstract

Traffic signs have an important role in ensuring the safety and smooth flow of driving activities. However, there are still many drivers who do not pay attention to traffic signs, either due to lack of focus, high speed, or environmental conditions that reduce visibility, thereby increasing the risk of road accidents. The purpose of this research is to develop a real-time traffic sign detection system that can provide immediate warnings to drivers. This system implements YOLOv10 to detect objects quickly and accurately, even in complex environmental conditions such as low lighting or varying angles of view. Text-to-Speech (TTS) technology is integrated so the system can convert detection results into voice alerts, allowing information to be received without distracting the driver's concentration. The traffic sign images used consist of 22 traffic sign classes obtained from open-source platforms like Roboflow and Kaggle. Model training was conducted using Google Colab with optimal parameters: a batch size of 8 and image resolution of 320×320 pixels. The trained model was then implemented into an Androidbased mobile application and is capable of real-time detection with an inference time of 300–600 ms. Testing results using 48 MP and 50 MP cameras showed a precision rate of 96%, recall of 93%, and overall accuracy reaching 90%. This system has been successfully implemented on Android devices and is capable of detecting and providing real-time voice alerts for traffic signs.