NICU Warmer — Intelligent Patient Monitoring#
Note! This application is for reference and evaluation purposes. It is not intended for direct use in clinical or diagnostic environments and is not validated as such.
The NICU Warmer application is a reference solution that demonstrates how multiple AI models can run simultaneously in a single GStreamer pipeline on Intel® hardware, providing workloads that mimic real-time neonatal patient monitoring in a simulated hospital warmer bed scenario.
It combines several representative AI workloads:
Object Detection (×3): Custom OpenVINO FP32 models for detecting patient presence, caretaker presence, and warmer latch clip status — all running on Intel Arc GPU.
rPPG (Remote Photoplethysmography): Contactless heart and respiratory rate estimation from facial video using MTTS-CAN, running on CPU.
Action Recognition: Kinetics-400 encoder/decoder model mapped to 11 NICU-specific activity categories, running on Intel NPU (AI Boost).
Metrics Collector: Gathers hardware and system telemetry (CPU, GPU, NPU, memory, power) from the host.
UI: Web-based React dashboard for visualizing detections, vital signs, activity, and system performance in real time.
Together, these components illustrate how vision-based AI workloads can be orchestrated across Intel GPU, NPU, and CPU, monitored, and visualized in a clinical-style scenario.
Supporting Resources#
Get Started – Step-by-step instructions to build and run the application using
makeand Docker.System Requirements – Hardware, software, and network requirements, plus an overview of the AI models used by each workload.
How It Works – High-level architecture, service responsibilities, and data/control flows.
Release Notes – Version history and known issues.