Real-Time Mobile App Traffic Sign Recognition With Yolov10 and CNN for Driving Education

This study presents a novel Traffic Sign Recognition system for Android devices, employing Convolutional Neural Networks (CNNs) and the YOLOv10 architecture for real-time detection and classification of Philippine traffic signs. The application improves road safety by providing auditory and visual cues for traffic sign compliance, especially in the context of driving education. The system integrates TensorFlow Lite (TFLite) to optimize performance for resource-constrained mobile platforms. The study encompasses data collection, annotation, preprocessing, model development, hyperparameter tuning, model training, model evaluation, and application development. The detection model achieved high accuracy with a mean Average Precision (mAP) of 0.823 and 99.66% accuracy for the classification model. The developed mobile app also demonstrated effective real-time recognition capabilities with a recognition inference time of 200-300ms. Challenges such as low-light performance are identified, with recommendations for future enhancements in data balancing, nighttime functionality, and multilingual feedback. This scalable, cost-effective system bridges the accessibility gap in advanced driver assistance technologies, offering the potential for wider regional adaptation.