Semantic Structure From Motion for Depth Cameras (ros_sfm)

Structure From Motion for ROS(noetic). Uses pcl and standard ros packages. Uses Libtorch and OpenCV for semantic segmentation.

⚙️ Setup

Dependeces (ROS, PCL, pcl_conversions, pcl_msgs, Libtorch, OpenCV)

Libtorch

Visit Offical Libtorch installation guide.

Package

1. Clone the GitHub repository into 'catkin_ws/src'

git clone https://github.com/jagennath-hari/ros_sfm.git

2. 'catkin build -j -DCMAKE_BUILD_TYPE=Release' and source the workspace.

🖼 Demo

Download bag file and paste inside 'bag' folder in sfm. To run a demo.

roslaunch sfm structure_from_motion_example.launch

🏁 To use

roslaunch sfm semantic_sfm.launch rgb_topic:=/rgb_topic depth_topic:=/depth_topic camera_topic:=/camera_topic odom_topic:=/odom_topic model_path:=/path leaf_size:=/double

Topic inputs

• rgb_topic(sensor_msgs/Image)
• depth_topic(sensor_msgs/Image)
• camera_topic(sensor_msgs/CameraInfo)
• odom_topic(nav_msgs/Odometry)

Params

• model_path (string)
• leaf_size (double)

📊 Visualization

Rviz can be used to view the topics '/sfm/cloud' and '/sfm/trajectory'.

Published topics

• /sfm/cloud(sensor_msgs/PointCloud2) (To view semantic point cloud switch from RGB8 to Intensity and select channel label)
• /sfm/trajectory(nav_msgs/Path)
• /sfm/MapGraph (sfm/MapGraph)