Surface Pro 5 IR Camera Pipeline (Rust)

A high-performance, stable 30FPS IR camera pipeline with hardware-synchronized IR LED control for the Microsoft Surface Pro 5 (OV7251 sensor). This project is designed specifically to bring stable, bright IR video to Linux, enabling facial recognition tools like Howdy.

Features

• Stable 30FPS: Normalizes sensor vertical blanking to guarantee a stutter-free, fluid video stream.
• Hardware IR LED Control: Bypasses driver limitations to manually strobe the IR LED via direct I2C injection.
• Super-Boost Brightness: Dynamically scales sensor exposure lines (up to 1500) and analog gain (up to 8x) for clear imagery even in pitch darkness.
• Low-Latency ISP: A dedicated Rust unpacker that converts Intel IPU3 10-bit raw packed frames into standard BGRA format in real-time.
• Universal compatibility: Outputs to a v4l2loopback virtual device (/dev/video20), making the IR camera usable by any standard Linux video application (ffplay, fswebcam, browsers).

Prerequisites

This pipeline requires a Linux environment (tested heavily on Fedora with the linux-surface kernel). Install the necessary dependencies:

sudo dnf install rust cargo v4l-utils i2c-tools v4l2loopback

Ensure the v4l2loopback kernel module is loaded:

sudo modprobe v4l2loopback devices=1 video_nr=20 card_label="Surface IR Camera" exclusive_caps=1

Installation

1. Clone the repository:

git clone https://github.com/Yourumark/surface-IR-rust.git
cd surface-IR-rust

2. Build the main ISP and the injector daemon:

# Build the main ISP (raw frame unpacker)
cargo build --release

# Build the I2C scanner/injector daemon
cd scanner
cargo build --release
cd ..

Usage

Start the pipeline using the provided shell script. You can pass a brightness level between 0 and 255.

• 64 is recommended for standard indoor use.
• 180 activates "Super-Boost" for dark environments.

./run_capture.sh 180

Once running, you can view the live stream:

ffplay -f video4linux2 -pixel_format bgr0 -video_size 640x480 -i /dev/video20

How It Works (The Technical Challenge)

See ARCHITECTURE.md

License

This project is licensed under the MIT License - see the LICENSE file for details.

Dependencies & Acknowledgments

This project heavily relies on the following open-source tools. We respect their respective licenses and thank the maintainers:

• v4l-utils (GPL-2.0)
• i2c-tools (GPL-2.0)
• v4l2loopback (GPL-2.0)
• Built with the Rust v4l crate (MIT/Apache-2.0)