Paper Title
REAL-TIME ROAD SIGN DETECTION AND IDENTIFICATION ON MOBILE DEVICES

Abstract
Road sign detection and identification are critical components of intelligent transportation systems and advanced driver-assistance applications. This paper presents a real-time road sign detection framework designed for deployment on Android mobile edge devices. The proposed system employs the YOLOv11 object detection architecture [1], optimized via TensorFlow Lite (TFLite) quantization for efficient on-device inference. A key focus of this work is the robust detection of partially occluded and perspective-distorted road signs, which frequently appear in real-world driving environments. The model is trained on a custom-annotated dataset comprising 8,577 images across 68 Indian road sign classes, with 6,082 total annotated bounding-box instances. The best-performing Float32 variant achieves a mean average precision (mAP@50) of 78.6% and mAP@50-95 of 67.7%, a precision of 81.7%, and a recall of 75.1%, while sustaining 12.5 FPS on a Snapdragon 870-based Android device. These results confirm that high-performance deep learning models can be effectively deployed on resource-constrained mobile platforms without significant accuracy degra-dation. Keywords - Road Sign Detection, Yolov11, Tensorflow Lite, Edge Computing, Object Detection, Mobile Vision, Autonomous Driving, Indian Traffic Signs