ROS2 Camera Publisher for Raspberry Pi

A lightweight, optimized ROS2 node for publishing compressed camera images from Raspberry Pi - solving the v4l2_camera dual-publishing overhead problem.

Show Image Show Image Show Image

Problem Solved

The Issue: The standard v4l2_camera node publishes BOTH raw and compressed images simultaneously, which:

• Wastes CPU resources (double encoding)
• Consumes unnecessary bandwidth
• Causes performance bottlenecks on Raspberry Pi
• Results in dropped frames and lag

The Solution: This package publishes only compressed JPEG images , providing:

• 50% less CPU usage
• Consistent 30 FPS with no lag
• Lower memory footprint
• Optimized specifically for Raspberry Pi 4B ARM architecture

✨ Features

• Compressed-only publishing - no raw image overhead
• ARM Cortex-A72 optimizations - compiler flags tuned for Pi 4B
• Configurable parameters - resolution, framerate, JPEG quality
• Optional republisher - convert to raw when needed (separate launch)
• Zero frame drops - tested stable at 30 FPS
• Low latency - efficient JPEG encoding with OpenCV

Requirements

• ROS2 Humble
• Raspberry Pi 4B (or compatible ARM board)
• OpenCV (libopencv-dev)
• USB Camera or Pi Camera Module (with V4L2 driver)
• Optional: image_transport for raw republishing

Installation

bash

# Navigate to your ROS2 workspace
cd ~/ros2_ws/src

# Clone the repository
git clone https://github.com/yourusername/rpi_camera_publisher.git camera_publisher

# Install dependencies
sudoapt-getinstall libopencv-dev ros-humble-image-transport

# Build with optimizations
cd ~/ros2_ws
colcon build --packages-select camera_publisher --cmake-args -DCMAKE_BUILD_TYPE=Release

# Source the workspace
source install/setup.bash

Usage

Usage

bash

# Run with default settings (640x480 @ 30fps)
ros2 run camera_publisher camera_node

# Custom parameters
ros2 run camera_publisher camera_node --ros-args \
  -p frame_rate:=30\
  -p jpeg_quality:=75\
  -p image_width:=1280\
  -p image_height:=720\
  -p camera_id:=0
# Raw camera (decompressed and can be viewed directly from Rviz2)
ros2 launch camera_publisher camera_pub.launch.py

This launch file includes:

• camera_node - Publishes compressed images
• camera_republisher - Converts compressed to raw

Performance Benchmarks

Tested on Raspberry Pi 4B (8GB RAM, Ubuntu 22.04 server LTS) :

Resolution
FPS
CPU Usage
Status
640x480
30
~12-15%
✅ No lag
1280x720
30
~22-25%
✅ Stable
1920x1080
15
~28-32%
✅ Smooth

Comparison with v4l2_camera_node:

• 🔴 v4l2_camera: 30-40% CPU @ 640x480 (publishes raw + compressed)
• 🟢 This package: 12-15% CPU @ 640x480 (compressed only)

Configuration Parameters

Parameter
Type
Default
Description
frame_rate
int
30
Frames per second
camera_id
int
0
Camera device ID (/dev/videoX)
jpeg_quality
int
80
JPEG compression (0-100)
image_width
int
640
Image width in pixels
image_height
int
480
Image height in pixels
frame_id
string
"camera"
TF frame ID

Published Topics

Topic
Type
Description
/camera/image/compressed
sensor_msgs/CompressedImage
Compressed JPEG image

When using the republisher launch file:

Topic
Type
Description
/camera
sensor_msgs/Image
Raw image (on demand)
/camera/camera_info
sensor_msgs/CameraInfo
Camera parameters

Project Structure

camera_publisher/
├── CMakeLists.txt          # Build configuration with ARM optimizations
├── package.xml             # ROS2 package manifest
├── LICENSE                 # MIT License
├── README.md               # This file
├── src/
│   └── main.cpp            # Camera node implementation
├── launch/
│   └── camera.launch.py    # Launch file with optional republisher
└── include/
    └── camera_publisher/   # (empty, reserved for headers)

Optimization Details

The CMakeLists.txt includes ARM-specific optimizations:

• -march=armv8-a+crc+simd - ARM Cortex-A72 with SIMD
• -mtune=cortex-a72 - CPU-specific tuning
• -O3 - Maximum optimization level
• -ftree-vectorize - Auto-vectorization for loops
• -ffast-math - Aggressive math optimizations

These flags provide ~20-30% performance improvement on Raspberry Pi 4B.

Troubleshooting

Camera not detected:

# List available cameras
ls /dev/video*
v4l2-ctl --list-devices

# Test camera with OpenCV
python3 -c "import cv2; print(cv2.VideoCapture(0).read())"

Permission denied on /dev/video0:

sudousermod -a -G video $USER
# Logout and login again

Contact

For questions or suggestions, please open an issue on GitHub.

---