sgm_gpu is a ROS2 package based on Semi-Global Matching on the GPU by D. Hernandez-Juarez .

It contains library, component and node to estimate disparity image from stereo images using NVIDIA GPU.

Prerequisite: ROS2 Foxy and CUDA in Ubuntu 20.04

1. Create a workspace for colcon

2. Install dependencies (See package.xml)

3. Put sources in the workspace

   cd <the workspace>
   git clone https://github.com/fujimo-t/ros2_sgm_gpu.git src/sgm_gpu

4. Build

   colcon build
   

ros2 launch sgm_gpu sgm_gpu_test.py

Then viewers for input stereo images and output disparity are opened like above screenshot.

Estimate and publish disparity image from stereo image.

• ~/disparity (stereo_msgs/DisparityImage)

Remap them to topics from stereo camera.

• left_image (sensor_msgs/Image)
• right_image (sensor_msgs/Image)
• left_camera_info (sensor_msgs/CameraInfo)
• right_camera_info (sensor_msgs/CameraInfo)

• image_transport (string, default: "raw")

  See image_transport's documentation.

Component version of the node.