robot_urdf_filter

A package that prepares images and pose data for 6D pose tracking.

Data directory

/depth_data: original depth images (one image for each frame), joint positions and joint velocities from the rosbag, real depth from depth images and simulated depth by drake as txt files.

/rgb_data: original rgb images from the rosbag.

/mask_data: robot masks generated by urdf_filter.

/dilated_mask_data: dilated masks of the robot by increasing the margion of masks by 3 pixels.

/filtered_data: screen images and depth images without the robot generated by urdf_filter.

/cube_data: screen images and depth images of the cube generated by depth_filter.

/denoise_cube_data: final denoised depth images of the cube.

/tagslam_poses: 4 by 4 transformation matrix of the object retrieved from odom.bag and timestamps.txt for duration calculation.

Data generation procedure

• Generate depth data: In rosbag_processor.py, run bag_to_depth_images() to extract depth images from rosbag and write images.txt
• Generate rgb data and joint positions: In rosbag_processor.py, run extract_poses_with_timestamps() to extract joint positions and rgb images at matched timestamps with depth topic from rosbag
• In file_utils.py, run write_real_depth_as_txt() to save real depth txt at once
• Generate robot masks: In urdf_filter.py, run main() to get robot masks and filtered depth images
• Generate dilated robot masks: In urdf_filter.py, run dilate() to get dilated robot masks.
• Generate depth and rgb data without robot: In file_utils.py, run main() to save filter depth and rgb images in filtered_data.
• Generate cube depth images and screen images: In depth_filter, run main() to get cube screen images and depth images
• Denoise cube depth images by running denoise() in file_utils.py.
• Calculate cube initial pose: In rosbag_processor.py, run extract_cube_pose() with desired start_frame to get the initial cube pose in camera frame and save in BundleTrack annotated_poses directory.