“KUBERNETES–MANAGED EDGE–BASED AI SYSTEM ASSISTING PERSONS WITH DISABILITIES“

Abstract

This study examines the design, implementation, and assessment of RealTimeCam, a lightweight edge computing system that integrates IoT devices with Kubernetes orchestration for real–time AI inference. The system utilizes a heterogeneous cluster comprising a virtualized master node operating on Ubuntu 22.04 and ARM–based Raspberry Pi worker nodes, integrating virtualized control with physical edge computation. K3s is a lightweight Kubernetes distribution that orchestrates containerized workloads, such as YOLOv8 for object detection and ByteTrack for multi–object tracking, while MediaMTX facilitates low–latency RTSP/WebRTC video streaming. Performance evaluations indicate efficient CPU and memory usage, stable thermal and energy profiles, and near real–time pod scheduling across diverse nodes. Network analysis demonstrates latency below 10 milliseconds and consistent bandwidth allocation for real–time video and AI applications. A comparative analysis reveals the benefits of open–source, cost–effective, and energy–efficient edge deployments in relation to current IoT–edge frameworks. The discussion addresses limitations related to hardware constraints, ephemeral storage, and single–master topology. Future directions encompass multi–node scaling, GPU/NPU acceleration, federated learning, and dynamic workload management.