Lightweight, Low-Cost Embedded Radar-Gated Camera Architecture for MCU-Level Motion-Triggered Occupancy Sensing

Abstract

A low-cost embedded radar–camera fusion architecture is presented for MCU-level motion-triggered occupancy sensing in practical environments. Unlike vision-only approaches that are vulnerable to illumination variation and occlusion, and radar-only approaches that suffer from clutter-induced false alarms, multipath responses, and intermittent target updates, the proposed architecture adopts a radar-centric hierarchical fusion strategy for resource-constrained embedded platforms. A 24 GHz FMCW radar serves as the primary event trigger, and an RGB camera is activated only for scene-change verification; occupancy is asserted only when radar-detected motion and camera-observed scene variation are mutually consistent. The sensing target is motion-triggered occupancy, rather than static presence monitoring, physiological micro-motion sensing, or semantic human recognition. Experimental verification is performed on an ESP32-S3-based embedded platform under day/night conditions and front-facing, diagonal, and back-facing motion. Compared with radar-only and camera-only baselines, the proposed fusion architecture improves recall by up to 16 percentage points and precision by up to 9 percentage points, achieving event-level F1-scores of up to 92% under daylight and 90% under nightlight. Temporal stability analysis shows that the fusion output maintains occupancy for approximately 90% or more of the true presence duration in the front-facing and diagonal scenarios and reduces false ON/OFF toggling by up to 58.4% relative to radar-only operation. Ablation analysis confirms that radar-gated execution reduces the camera-decision duty cycle to 7.76% in the evaluated sequence, and parameter sensitivity analysis clarifies the trade-off between temporal stability and responsiveness. Real-time implementation results verify the practicality of the pipeline for low-cost MCU-level occupancy sensing.