Design and development of a TinyML-Based Intelligent Home Security System using computer vision and deep learning. (Case study: Intruder Detection)

Abstract

With rising security concerns in both urban and rural areas, the need for intelligent, real-time, and cost- effective home security systems has become increasingly urgent. Traditional approaches such as hiring security personnel or deploying basic surveillance cameras often fall short due to their reliance on human monitoring and delayed response. To address these challenges, this study presents the design and implementation of a TinyML-based intelligent home security system that leverages computer vision and deep learning to detect intrusions in real time on resource-constrained edge devices. The proposed system combines face recognition and behavior detection to verify the identity of individuals and assess their actions. It employs a Convolutional Neural Network (CNN) for facial recognition and YOLOv8, a state-of-the-art object detection model, to identify suspicious behaviors such as wearing a mask or carrying a weapon. These models operate jointly to make robust classification decisions. The system is deployed on a Raspberry Pi 5, integrated with Passive Infrared (PIR) and Light Dependent Resistor (LDR) sensors to respond intelligently to motion and ambient lighting changes. When an unknown or suspicious individual is detected, an alert is triggered, and real-time feedback is provided to the homeowner. The dataset used for training and evaluation consists of 1,100 labeled images, including both real and synthetically generated data. The integrated model achieved an overall accuracy of 97.27%, with precision, recall, and F1-score values exceeding 96%, demonstrating the system’s strong classification performance. Despite these results, the system encountered limitations in extreme low-light conditions and when intruders wore facial coverings that resembled authorized users. These limitations highlight areas for future research and system enhancement. Overall, this work demonstrates the feasibility of deploying lightweight, privacy-preserving security solutions using TinyML and embedded AI, offering an efficient and reliable alternative for modern home security applications.