Simple implementation of Dense Visual Odometry with RGB-D camera

This ROS package provides simple implementation of Dense Visual Odometry with RGB-D camera (https://vision.in.tum.de/_media/spezial/bib/kerl13icra.pdf&hl=ja&sa=X&scisig=AAGBfm147Qu9xs7R6FGEiy3zbmEXLYgvbw&nossl=1&oi=scholarr).
Dataset is available from https://vision.in.tum.de/data/datasets/rgbd-dataset/download# .
This implementation is just for dense tracking. Robust weighting is not implemented yet.

Dependency

Ubuntu 14.04 LTS or 16.04
ROS indigo/kinetic
OpenCV
PointCloudLibrary

Example

• Dowload sample rosbag file https://vision.in.tum.de/rgbd/dataset/freiburg2/rgbd_dataset_freiburg2_desk.tgz
• Run rosrun simple_dvo main /camera/rgb/image_color /camera/depth/image /camera/rgb/camera_info
• Runrosbag play rgbd_dataset_freiburg2_desk.bag
• Run rosrun rviz rviz
• In rviz "Fixed Frame" tab, change "world" to "cam_origin"
• "Add" -> "By topic" and add "PointCloud2"
• "Add" -> "By display type" and add "TF"

You can see the camera pose and pointcloud in rviz.

References

• Robust Odometry Estimation for RGB-D Cameras (C. Kerl, J. Sturm, D. Cremers), In Proc. of the IEEE Int. Conf. on Robotics and Automation (ICRA), 2013
• Dense Visual Odometry (https://github.com/tum-vision/dvo)