Edge Artificial Intelligence

13AIPendant: An Efficient and Lightweight Threat Detection System for Real-Time Personal Safety

Sanghyun Yeo¹*, Vu Minh Phuc² and Le Anh Ngoc³

¹Global Leadership College, Bachelor of Engineering, Yonsei University, Seoul, South Korea

²BDA, Vietnam National University, Hanoi, Vietnam

³Swinburne Innovation Lab, Swinburne Vietnam, FPT University, Hanoi, Vietnam

Abstract

This paper presents the development and evaluation of a novel AI Pendant system for real-time threat detection and personal safety. The system uses MobileNetV2, a lightweight convolutional neural network, to identify harmful objects such as guns and knives. The model is trained on the HOD Benchmark Dataset and achieves a validation accuracy of 72.2%, demonstrating strong performance with minimal computational requirements. The system integrates with Telegram, a free messaging API, to deliver real-time SOS alerts when a threat is detected. Our results demonstrate the model’s suitability for deployment on resource-constrained devices, offering a cost-effective and scalable solution for personal safety.

Keywords: Threat detection, MobileNetV2, lightweight AI, personal safety, SOS alert, real-time detection, HOD dataset

13.1 Introduction

13.1.1 Background

The increasing prevalence of violent crimes highlights the urgent need for portable, real-time safety devices. While conventional safety systems rely on computationally expensive object detection models, these are unsuitable for resource-constrained applications ...