NVIDIA-Accelerated FLIR ONE Pro Edge Pipeline (UW Medicine)

An enterprise-grade, edge-optimized medical research pipeline designed to capture high-fidelity radiometric thermal streams from the FLIR ONE Pro (Lepton 3.5 Core). This system establishes a hardware-accelerated processing engine on NVIDIA RTX 6000 Ada workstations and NVIDIA Jetson Orin edge units to extract true kelvin/celsius matrices, handle co-registration with visible MSX feeds, and perform zero-copy VRAM spatial delta tracking.

Developed for the UW Medicine Clinical Imaging & Research Network.

Architecture Ecosystem Overview

The platform implements an asynchronous, multi-threaded worker pipeline to ingest mobile thermal arrays without frame dropping, maintaining a stable 8.7 Hz execution rate (FLIR regulatory ceiling):

1. Hardware Ingestion Layer (flir_stream_ingest.py): Utilizes low-level libusb/pyusb endpoints to bypass proprietary mobile applications. It establishes direct raw byte-stream communication to extract the interleaved MSX visible (640x480) and raw radiometric thermal (160x120) sub-frames.
2. NVIDIA CUDA Thermal Core (cuda_thermal_math.py): Shifts raw 14-bit linear ADC counts directly onto GPU registers. Using compiled CUDA kernels, it executes real-time radiometric calibration calculations (Planck constant corrections, ambient temperature, distance, and emissivity scaling) across the entire frame matrix simultaneously.
3. Anatomical Co-Registration Layer (ocr_layer.py / pipelines/): Utilizes NVIDIA TensorRT-optimized models to dynamically warp and map the lower-resolution thermal map perfectly onto the high-resolution visible matrix, accounting for parallax offsets.
4. Clinical Telemetry Hub (stream_dashboard.py): Forwards localized region-of-interest (ROI) max/min delta points to a secure internal web server running strictly inside the UW Medicine intranet.

Host Prerequisites

Before deploying the pipeline container stacks, ensure your clinical workstation or Jetson edge node matches the hardware baseline:

1. NVIDIA Compute Architecture: Driver version 550.x or newer with CUDA Toolkit 12.2+ integrated.
2. Access Permissions: Local udev rules mapped to permit containerized USB access to the FLIR vendor ID.
3. Containerization: Native Docker runtime setup paired with the NVIDIA Container Toolkit.

# Add FLIR ONE Pro hardware communication mapping to host udev rules
echo 'SUBSYSTEM=="usb", ATTR{idVendor}=09cb, ATTR{idProduct}=1996, MODE="0666", GROUP="plugdev"' | sudo tee /etc/udev/rules.d/99-flir.rules
sudo udevadm control --reload-rules && sudo udevadm trigger

Rapid Clinical Deployment

1. Build & Spin Up the Edge Node Containers

The framework includes dedicated optimization paths for both enterprise desktop setups and mobile Jetson carts.

# Clone the repository structure
git clone https://github.com
cd NVIDIA-Accelerated-FLIR-ONE-Pro-UWMed

# Build and run the local container stack via Docker Compose
docker compose up --build -d

2. Verify Hardware Pipeline Connection

# Tail live container logs to verify raw radiometric frame ingestion loop
docker exec -it flir_nvidia_worker tail -f /var/log/thermal_pipeline.log

HIPAA Compliance & Data Security Protocol

To maintain strict alignment with UW Medicine Security Regulations and HIPAA Safe Harbor Laws, this pipeline operates under rigid data isolation rules:

• Zero PHI Retention: The acquisition layer blocks text inputs for patient identifiers. All files are sequentially indexed utilizing automated UUID hashes.
• VRAM Volumetric Anonymization: Visible spectrum components mapped through the MSX alignment engine undergo automatic spatial blurring across high-risk facial features on the GPU before saving.
• Network Isolation: Edge components bind exclusively to internal standard enterprise networks (10.0.0.0/8). Cloud transmission of raw files is locked behind a TLS 1.3 encryption wrapper.

Clinical Production Disclaimer

The FLIR ONE Pro is classified as a prosumer thermal device. Telethermographic systems are regulated by the FDA as Class I devices under 21 CFR 884.2980. This pipeline is engineered exclusively for adjunctive clinical trial support, laboratory tracking, and institutional research. It is strictly unauthorized and uncleared for use as a standalone diagnostic diagnostic tool.