Stephen Meschke's practice log

Stephen Meschke#

vision research, 1 minute

Working on creating a juggling dataset.

Creating a juggling dataset was my original goal, but it kinda morphed into this project: https://www.reddit.com/r/juggling/comments/6aa9ov/juggling_pattern_waveforms/

There was some interest in the data that I used to create the graphs on the video, which instilled hope in me that there is some interest in a juggling data set.

What is a juggling data set anyway?

* The data set consists of several clips of juggling (about 600 frames long each)
* Each clip of juggling has an associated .csv file that contains the locations of each ball in each frame.
* The data gets 'big' pretty fast. For a 10 second clip with 5 balls there are 5*10s*60fps = 3000 observations

Alongside juggling, I have also been training computer vision. I have created/learned to use some programs that can, with a little help, track the locations of the balls. Under optimal conditions, I can track with 100% accuracy fully automated. Under good conditions, I can track at 98% accuracy with 95% automation.

Optimizing the conditions before filming juggling is not difficult; have good lighting and a high degree of contrast between the balls and the other objects in the frame. It helps to use different brightly colored balls too.

Tracking 5 balls for 600 frames took me about 10 minutes. With this level of automation, it has become practical to start building a data set.

Here is a link to the data that I have created so far: https://drive.google.com/drive/folders/0B7QqDexrxSfwcmROb1ByNkhqOEU?usp=sharing

I would like to introduce this idea with a post in small talk, but I am not sure how to go about that. I estimate that less than 50% of jugglingedge.com users are really into data science.

Total practice time: 1 minutes

Location: Linux

Comments (2)
Orinoco - Never worry about potential audience size. My lower limit for the number of people who I think will be interested in something before I post is one! I think this is fascinating & I know several others think so too.
Stephen Meschke - Thanks. I am want to polish the presentation a little bit, and make some tutorials that explain how to make the graphs in a spreadsheet. I am excited about this and looking forward to the responses!