MOTOGUARD-AI: Real-Time Motorcycle Theft Detection Using YOLO Architecture

Abstract

Motorcycle theft remains a serious and persistent security problem with conventional monitoring systems. To address this challenge, this study proposes MOTOGUARD-AI, a real-time motorcycle theft detection framework based on the YOLOv12 architecture that integrates motorcycle, license plate, and rider face detection. The system automatically triggers dual-channel alerts via WhatsApp (Twilio API) and email (SMTP) when ownership mismatches or suspicious riding behavior are detected, delivering visual evidence and contextual information within seconds. The model was trained and evaluated on a custom dataset of 10,000 annotated images and video frames collected from Indonesian urban CCTV footage. Experimental results show that MOTOGUARD-AI achieves 92.5% [email protected] and 76.8% [email protected]:0.95, with an average inference latency of 12 ms on edge GPUs and an end-to-end detection-to-notification time of under 3 seconds, outperforming YOLOv11-based baselines in both accuracy and robustness. The system also attains 88.6% theft indication accuracy and over 99% notification delivery success, demonstrating that the attention-centric design of YOLOv12 significantly improves small-object and occlusion handling while enabling a practical, scalable, and proactive anti-theft solution for smart city surveillance in Indonesia.