OpenPose: Multi-person Tracking and Feature Extraction

## Expected Behavior:

1. Track 'who is who' given openpose output where multiple people are detected, creating a separate csv for each person
2. Data visualization(s) supporting #1, to help determine which csv belongs to which participant
3. Update estimates of openpose keypoint locations (likely method: using kalman filtering to predict locations based on previous locations- currently openpose does frame-by-frame estimations only)
4. Track optical flow in the area of each person. This is an additional feature describing movement of each detected person

**Note: this issue corresponds to branch 'openpose'**

Start from the `openpose` branch and inspect how OpenPose output is currently processed and stored. Break the work into multi-person tracking, per-person CSVs, visualizations, keypoint prediction, and optical-flow features; done means each requested behavior is implemented and the resulting data identifies participants.

Written by the indexing model from the issue text.