A Low-Cost, Weather-Adaptive, High-FPS Multi-Sensor Edge Surveillance Node

Abstract

Outdoor surveillance fails exactly when it is needed most-in rain, fog, darkness, and fast motion-and the usual fixes (always-on high frame rate, cloud inference) are power-hungry and expensive. This paper presents AWASN, an all-weather adaptive surveillance node that fuses a 4K RGB camera, dual-band infrared, and a PIR motion sensor on a single Rockchip RV1126 edge platform, classifies the weather ondevice, and reconfigures imaging, frame rate, and power accordingly. The contribution is the integrated, measured design. An on-device weather classifier reaches a cross-validated 86.5% accuracy across day/night/rain/fog (fog hardest, as expected from its confusion with rain), and a sensor-fusion ablation shows each modality matters: RGB alone gives 63.9%, infrared raises it to 75.6%, and adding PIR reaches 86.7%—a 23-point gain from fusion. An optical-flow-driven controller adapts the frame rate (15-120 FPS), cutting average power 63% versus always-on operation (1.8 W vs 4.87 W) and extending battery life, while a 120 -FPS, 50 - $\mu \mathrm{s}$-shutter mode reduces motion blur from 18.4 px to under 0.1 px at 200 km/h. Against Jetson Nano, Coral, and Raspberry Pi baselines, AWASN delivers competitive accuracy at twice the frame rate, roughly half the power, and a USD 83 bill of materials. With the hardware resolved, the models specified, and the subsystems quantified, AWASN is a verifiable low-power edge surveillance platform.