data_acquisition

Data acquisition

We recorded the data using the duckietown localization system

To record data follow the instruction here. We recommend to increase the buffer size of rosbag record with the -b tag (we used -b 2048) and to split the recording into smaller ones (we used 15 seconds split) using either the --split tag, e.g.,

$ rosbag record --split --duration=10 -b 2048 -a

or after the recording with the split bash script we provide. Increasing the buffer size avoids the dropping of messages during the recording with the localization system and splitting the rosbag will lower the memory usage of the post-processing and optimization.

To automate the post-processing and optimization we provide two functions post_process.sh and optimize.sh. For details see Utilities.

Data Labeling

For details see data_labeling/README.md

Utilities

Utilities to automate the data acquisition

rosbag_split.sh

Split the recorded bag file into smaller ones. The lengths of the split can be defined in the bash file. Just change the variable STEPS to the desired time, e.g., STEPS=15 for a split duration of 15 seconds. To execute run

$ bash rosbag_split <BAGFILE>

where <BAGFIlE> is the name of the bag file without the .bag ending

post_process.sh

If the recorded data is split it's more convenient to use our post_process.sh script. The split data has to be stored in a single directory and have the following naming convention bagname_.bag, e.g., recording1_0.bag, recording1_1.bag etc. Execute the following command.

$ bash post_process <Host IP Address> <path/to/rosbag> <sequence start> <sequence end>

optimize.sh

If the post_processed data is split, it's more convenient to use our optimize.sh script. The split data has to be stored in a single directory and the have the following naming convention processed_bagname_.bag, e.g., processed_recording1_0.bag, processed_recording1_1.bag etc. Execute the following command.

$ bash optimize.sh <Host IP Address> <Host> <path/to/processed_rosbag> <sequence start> <sequence end>

You have to change the fork and map name in the bash file (variables FORK, MAP) to the desired fork and map.

process_image.py

ROS node that converts the rostopic /<VEHICLE_NAME>/imageSparse/compressed to jpg images, and stores the image name and the corresponding timestamp in a csv file. To use the python script change the variable path to the desired location of the jpg images and the csv file and the variable bot to the VEHICLE_NAME. The images are stored in a subfolder called images/ and have the naming convention <VEHICLE_NAME>_<TIMESTAMP_SEC>_<TIMESTAMP_NSEC>.jpg. The csv file is named image_timestamps.csv and has the format [<image name>, <sequence number image>, <timestamp seconds>, <timestamp nanoseconds>].

Example:

• change variable path = /home/oliolioli/recording/ and bot = autobot05
• Start a rosmaster
• Execute python script

python process_image.py

• Images and csv file are stored as follows:
  • /home/oliolioli/recording/
    • timestamps.csv
    • images/
      • autobot05_111111_111111.jpg
      • autobot05_111111_222222.jpg
      • ...