GST Bridge - GStreamer Camera Bridge

High-performance RTSP to MJPEG bridge using GStreamer with hardware acceleration support.

Features

• Hardware Acceleration: Auto-detects NVIDIA, VAAPI (Intel/AMD), or falls back to software decode
• Auto-Reconnection: Automatic 5-second reconnection on stream failures
• Low Resource Usage: <5% CPU per camera with hardware acceleration
• Scalable: Designed for 200-500+ cameras per server
• Production Ready: Systemd integration, robust error handling

Quick Start

Installation

Arch Linux / CachyOS:

cd pkg/arch
makepkg -si

Ubuntu 24.04:

./build/build_deb_package.sh
sudo dpkg -i dist/*.deb

Running a Single Camera

# Manual execution
RTSP_URL=rtsp://camera/stream PORT=8000 gst-bridge

# With systemd
sudo cp examples/camera.env /etc/koshee/cameras/frontdoor.env
# Edit /etc/koshee/cameras/frontdoor.env with your camera settings
sudo systemctl start gst-bridge-native@frontdoor

Multiple Cameras

# Scan network for cameras
gst-bridge-scan-cameras --network 192.168.1.0/24

# Generate configs
gst-bridge-convert-config cameras.json --systemd

# Start all cameras
/etc/koshee/cameras/start_all_cameras.sh

Documentation

• Hardware Acceleration - NVIDIA, Intel/AMD VAAPI, software decode
• GStreamer Scaling - Architecture and performance targets
• Systemd Services - Service configuration and management
• Build Guide - Building packages for different distributions
• Quick Start - Getting started quickly

Architecture

RTSP Camera → [Hardware Decode] → [Scale] → [JPEG Encode] → MJPEG/TCP Stream

Pipeline Variants:

• NVIDIA: nvh264dec → cudascale → nvjpegenc (full GPU)
• VAAPI: vah264dec → videoscale → jpegenc (Intel/AMD)
• Software: avdec_h264 → videoscale → jpegenc (CPU fallback)

Configuration

Camera configuration files in /etc/koshee/cameras/*.env:

RTSP_URL=rtsp://192.168.1.100/stream
PORT=8000
WIDTH=1120
QUALITY=90
# Optional: HW_ACCEL=vaapi|nvidia|software

See examples/camera.env for full example.

Building from Source

# Compile
make

# Run
RTSP_URL=rtsp://camera/stream PORT=8000 ./gst-bridge

Performance

Hardware	CPU Usage/Camera	Cameras/Server
NVIDIA GPU	2-4%	300-500+
Intel/AMD VAAPI	3-5%	200-400
Software Decode	15-25%	20-50

License

Apache-2.0

Author

Reid Nabinger reid@koshee.ai