25 July 2026 in SSRN Electronic Journal
MEMacarena Espinilla-Estévez
Macarena Espinilla-Estévez
The demographic shift towards an aging population demands intelligent Ambient Assisted Living (AAL) solutions; however, current monitoring systems typically force care facilities into a binary choice between invasive cloud video surveillance and context-limited traditional sensors. This paper proposes a novel, distributed edge computer vision architecture specifically designed for real-time, privacy-preserving indoor person detection and identification. Deployed on resource-constrained Raspberry Pi 4 edge nodes, the system integrates a YOLO11n detector and the SORT algorithm for spatial tracking, alongside InsightFace (ArcFace) for biometric resolution. To achieve functional real-time performance on ARM architectures, the pipeline leverages NCNN hardware acceleration and implements an opportunistic execution policy. By processing raw video locally and transmitting only serialized, anonymized JSON metadata and encrypted biometric embeddings to a centralized core server, the architecture follows a "privacy by design" approach and ensures GDPR compliance. Experimental evaluations across diverse dynamic scenarios demonstrate the system's biometric resilience under severe pose variations and partial occlusions, achieved through a robust multi-threshold matching strategy. Ultimately, this framework provides a scalable, low-cost, and non-invasive tool for continuous behavioral logging and anomaly detection in vulnerable environments.