Smart Home Security System Using Object Recognition with the EfficientDet Algorithm: A Real-Time Approach

Abstract

The EfficientDet method, which is implemented on the Raspberry Pi for real-time detection in resource-constrained contexts, is the basis for the smart home security system presented in this study.  The system integrates CCTV cameras, motion sensors, and detectors to identify and classify objects, sending notifications via WhatsApp via the Twilio API.  The EfficientDet-D0 model achieves an accuracy of 94.8%, an average processing time of 45 ms, and a memory usage of about 850 MB.  When compared to moving individuals or non-human things, testing shows that stationary human items have a higher detection accuracy.  Notifications are transmitted roughly every three seconds, with an average latency of 1.4 to 1.8 seconds.  The suggested method provides object recognition, real-time monitoring, and configuration flexibility in contrast to traditional IoT-based systems.  These results highlight the potential of EfficientDet as a reliable and adaptable solution for home security.  Future improvements include improving accuracy in a variety of environmental conditions and implementing adaptive learning.