Ros bag extractor

Extracts files from .bag files using ros nodes and rosbag python code API.

Note:

• Installation of ROS melodic is required!
• Requires ffmpeg for smartphone video frames extraction

Docker

Build the image:

docker build -t bag-extractor .

Note: you have to rebuild the image after updates in the source code

Run with:

docker run -v [LOCAL_PATH_TO_BAG]:/[BAG_FILENAME] -v [LOCAL_OUTPUT_DIR]:/output -t bag-extractor [BAG_FILENAME]

-v option is used to mount local directories to the image

Usage

Important: if you haven't edited your ~/.bashrc for automatic environment variable setup as suggested in ROS installation instructions 1.5 you need to run source /opt/ros/melodic/setup.bash in your current shell first.

---

On the first usage, run ./setup.sh to create virtual environment and install requirements.

1. Setup topic names for extraction in ./extract.conf file

2. Run ./extract.sh <PATH_TO_BAG> (optional)<PATH_TO_VIDEO> with path to your .bag file and path to smartphone video from OpenCamera Sensors as arguments. (note: directory with meta information about video, e.g. 20210119_233624, should be located in the same path as the video).

3. Preliminary data is saved to the subdirectories of ./output/<YOUR_BAG_NAME> directories (subdirectory name = topic name with / replaced with _). After this step the video is not split by frames to save space when storing the dataset.

4. Final extraction can be done locally by running ./local_extract.sh <DIR_WITH_EXTRACTED_DATA>. The script splits the video to frames and puts aligned timestamps as their filenames.

   • Optional: use --split as a second argument to get output for multiple sequences in subdirectories seq_0, ... ,seq_n

Final extracted structure:

File	Description
{camera_topic_name}	images from camera with timestamps as filenames in jpeg format
{depth_camera_topic_name}	depth images with timestamps as filenames in npy format (can be easily loaded to numpy with np.load())
{poincloud_topic_name}	point cloud files with timestamps as filename
{time_ref_topic_name}	time reference files, format: <sequence_number, timestamp, time_ref>
{imu_topic_name}	csv files, format: <timestamp, ox, oy, oz, ax, ay, az, temperature> where ox, oy, oz - angular velocity; ax, ay, az - linear acceleration
{camera_info_topic_name}	yaml files (can be easily loaded with yaml.load(data, Loader=yaml.SafeLoader in python)
smartphone_video_frames	extracted smartphone video frames with timestamps as filename

Samsung dataset timestamps alignment

Additionally, the tool aligns timestamps of the extracted data.

• Uses time reference topic to get time from another source, and writes transformation meta information to transformation_metainf.csv file in the image directory.

Currently alignment is implemented for:

• depth images
• smartphone images

Troubleshooting

• If you get this error: venv/bin/activate: line 57: PS1: unbound variable, you should update virtualenv version, see this stackoverflow answer for more info.