Identity Recognition Framework for Smart Door Automation Using YOLOv8 and Small Language Models (SLMs)

  1. Home
  2. March 2025
Sudip Chakraborty

Identity Recognition Framework for Smart Door Automation Using YOLOv8 and Small Language Models (SLMs)

Keywords : YOLOv8, Small Language Models, Multi-Agent Framework, Smart Door Automation, Identity Recognition, Home Automation Security, Real-Time Interaction, Embedded AI.


Abstract

In an era where home security meets smart technology, effective identity recognition has become pivotal, transforming traditional doors into intelligent gateways that offer both heightened security and personalized user interactions. This paper presents an Identity Recognition Agent integrated into a multi-agent smart door framework, employing YOLOv8 for real-time object detection and Small Language Models (SLMs) for generating dynamic, personalized natural language responses. Initially, all visitors are categorized as "Unknown," transitioning to "Known" upon repeated detections, with final classification into the Family category manually confirmed by the admin (Master). The proposed system utilizes a dual-stage recognition pipeline: YOLOv8 for rapid initial detection and a secondary face recognition model for precise identity verification, ensuring efficient, real-time identification. A persistent database tracks visitor detection frequencies, updating statuses automatically based on predefined thresholds. Compact, optimized SLMs generate nuanced, real-time voice responses tailored specifically to each visitor classification—Unknown, Known, Family, or Master. The architecture employs asynchronous, event-driven parallelism, significantly enhancing operational efficiency, latency reduction, and scalability. Evaluations affirm the system's robust real-time identity recognition, personalized interaction capabilities, and suitability for scalable deployment, marking a notable advancement in privacy-conscious, intelligent home automation solutions. The software architecture utilizes asynchronous event-driven parallelism for efficient performance and scalability, optimized through model quantization and hardware acceleration techniques. Evaluations demonstrate robust, low-latency identification and highly personalized, responsive interactions, marking significant progress in intelligent, privacy-preserving, and scalable home automation.

Download

Comments
No have any comment !
Leave a Comment