Create an AI-Driven Smart Home Security System

Part 3—Face Recognition and Package Detection

- By Dhairya Parikh

In the pre-Internet Dark Ages, all computing was done locally. Then storage and processing moved to the Cloud. Now, local models are gaining traction again. In this article, Pedro explores the fundamentals of edge computing. He explains the edge computing paradigm, compares it to the cloud-centric paradigm, then presents the advantages and challenges of edge computing, and some solutions in the IoT that could benefit from edge computing adoption.

In Parts 1 and 2 of this article series in the March and April, 2025 issues of Circuit Cellar [1] [2], we started building our AI-powered home security system with a Raspberry Pi 5 and a Hailo AI Kit. We gave it three key functionalities: The ability to detect people in real-time using YOLO (You Only Look Once) models running on the Hailo AI Kit. The model was configured to identify people with high confidence (which is customizable) and reduce false alarms. With an email alert system that notified us whenever an intruder was detected, using a Gmail account with two-step verification and app passwords. We can monitor the camera feed from any device on the same network and view detection results in real time via a local server using the Flask framework in Python to stream the security feed from the Raspberry Pi.

FIGURE 1 Web interface we created in the Part 2 of this series. (Yes, that is me!). The stream and detection updates are available across the local network.

IMPROVEMENTS

Our home security system is already capable of many things, such as detecting intruders, sending timely alerts, and providing real-time video feeds via a web interface (Figure 1). Although the current implementation is functional and adequate in some scenarios, a few gaps still need to be addressed to make the system truly reliable and user friendly.