Lightweight, Privacy-compliant Motion Detection System for Constrained Devices

In a rapidly evolving digital landscape, embedded vision systems are becoming increasingly prominent in applications such as home security, surveillance, and autonomous monitoring. However, existing solutions often lack computational
efficiency for embedded hardware or fail to meet key ethical and regulatory requirements. This paper presents a lightweight, privacy-compliant motion detection system with integrated facial recognition deployed on the Qualcomm RB5 platform. The system enforces data minimisation by recording video only when
motion is detected, reducing unnecessary data collection. To protect privacy, facial anonymisation is applied using Gaussian blurring, and Role-Based Access Control (RBAC) restricts access to sensitive data. Experimental evaluations demonstrate an
approximate 80% reduction in storage requirements compared to continuous recording while maintaining acceptable CPU and memory usage. The system also achieves reliable detection accuracy while aligning with data protection principles, making it suitable for privacy-sensitive, resource-constrained devices.