This repository contains files for processing OpenPose JSON output. Background for this work, including motivation and results, are written in detail in our paper which was presented at the International Learning Analytics and Knowledge (LAK) Conference 2022, and was nominated for the Best Short Paper Award.
Hur, P., & Bosch, N. (2022, March). "Tracking Individuals in Classroom Videos via Post-processing OpenPose Data". In LAK22: 12th International Learning Analytics and Knowledge Conference (pp. 465-471).
If video files should be concatenated before running OpenPose (e.g. one video is divided into many different video files), use make_ffmpeg_concat.py, which will create individual .bat files for the respective video files. The resulting .bat files will rely on the usage of FFmpeg to create concatenated .mov files.
Run OpenPose processing commands for your project video file using the 25-keypoint body/foot keypoint configuration, and saving the respective output keypoint files in JSON format to a directory on your computer.
Use concat_JSON_files.py to concatenate all JSON files into a formatted single 'FILENAME.csv' file with a descriptive header row. concat_JSON_files.py asks the user for the directory that holds JSON output, and then saves the concatenated file to the directory above the user-defined directory.
Tracking processing is completed through allprocesses_command.py, and arguments are handled through Python argparse. More details about possible arguments can be found toward the end of the script. Currently the "--only_track" argument must be used, as OpenPose processing code is yet to be integrated.
Processing will include calculating the closest distance (in pixels), closest distance owner (row), and second closest distance (in pixels), then subsequently tracking the skeletons and assigning IDs. In order to partially account for OpenPose's inconsistent detection of skeletons throughout frames.