GDPR and Juggling.

Search posts


Stephen Meschke -

GDPR and Juggling. Is recording a juggler without consent prohibited?

When the weather is nice, I love training outside. Occasionally, other park-goers will ask to record video of me for social media. I always give consent and then hit one of the quick tricks that I train for this specific scenario. This makes me feel good because it shows that juggling is interesting enough to record and share. I have been paid up to $5 as well!

I am 100% certain (but have no proof) that people record me without asking for my consent. This doesn't bother or intimidate me, but I can sympathize with other jugglers who are find this behavior intrusive and distracting. In the USA it is legal to record someone and then post that content online without their permission.

NPR reported that the GDPR, "forbids people from posting anyone's picture online without their permission." After hearing the news story, I am left with these questions:

  • Does the GDPR affect the legality of recording a juggler in public without their consent?
  • If a juggler finds images or videos taken and posted online without permission, what can be to get the content removed?
  • If filming a juggling video, it is necessary to obtain permission from people in the background before posting the video online?

Guili - - Parent

I've been recorded juggling many times, some of them without my consent... and it never bothered me. but i do think some people may feel it's not right. either for privacy, or for like trademark issues... you know. if someone records your show, you won't be able to sell a dvd later, some like that...
so, out of respect, yes, i think consent must be asked.

7b_wizard - - Parent

Don't spend those 5 bucks - keep 'em as a trophy!

Danny Colyer - - Parent

GDPR affects what businesses and organisations are allowed to do with personally identifying data, meaning data that can be used to identify an individual.

GDPR does not affect what individuals are allowed to do. In the UK, at least, there is no legal requirement to seek anyone's consent before taking photos or video in a public place, and that will continue to be the case.

I don't know what NPR is, or whether it's a reputable news source, but I'd hazard a guess that the story was written by someone who hasn't read GDPR. I have. Three times. And I wouldn't claim to to be certain about what it means in every possible data storage/processing/sharing scenario. But I can say with certainty that an individual would not be in breach of GDPR by posting a photo or video without the subject's consent. I suppose, conceivably, the company that owns the hosting site might have a problem. It's not something that's relevant to my reasons for reading the regulation, so it's not something I was looking for in any of my own readings, but I'm pretty sure there was no intent to ban photo sharing.

peterbone - - Parent

I think it's fine to film people in a public place as long as they remain anonymous. If you identify them when you publish the content then I think it would infringe GDPR.

Danny Colyer - - Parent

If you're publishing as an individual then you won't be in breach of GDPR. You can't be. GDPR doesn't cover actions by individuals not acting as part of an organisation.

Article 2(2)(c):

"This Regulation does not apply to the processing of personal data ... by a natural person in the course of a purely personal ... activity"

Stephen Meschke - - Parent


This is not related to the GDPR, but is related to juggling and data. Recent advances in computer vision have made extracting body pose and prop location quite easy. For example, this data was extracted from a video of yours: or

I'd like your opinion on this. Who does this data belong to, the person who created the source video or the person who extracted the data?

peterbone - - Parent

That's very interesting, especially as I'm a computer vision engineer. Do you have any more information about the method? In my opinion the data is public since I put the video online and therefore made it public. Any information extracted from that video is also public. I don't think it really belongs to anyone, although I think you have the right to keep the extracted data private if you wish.

Stephen Meschke - - Parent

Thank you for allowing me the use of the data.

Here is some more information about how I extracted the data from the video:

  • Optical flow was used to track both ends of both balance poles. This tracking was not fully automatic, it required user input 10 times.
  • The body pose was extracted using a neural network. A neural network is like a black box. An image of a person is put into one end of the box, and 15 coordinates describing the position of the head, neck, shoulders, elbows, wrists, etc...come out the other side of the box. This tracking was 100% automated. Link to the tutorial I learned from.
  • After I extracted the pose data, I took one additional step that really improved the results. The source video was 24 frames per second, so I had 24 sets of pose coordinates for every second of video. Using some simple algebra, I doubled the frame rate. I did this by calculating the midpoint between two poses in sequential frames. For example, if the position of the head in the first two frames for the source video is (100,100) and (110,110); the position of the head in the first three frames of the output video would then be: (100,100), (105,105), (110,110). This smooths the tracking results and also gives and odd lifelike quality to the stick-man.

peterbone - - Parent

Thanks. Interesting. I assume you downloaded the weights rather than training the NN yourself then? I was most impressed with the leg tracking as I'm wearing black baggy trousers in the video that tend to blend into one another. Let me know if you need any more video. I now have a camera that can film at higher frame rates.

Stephen Meschke - - Parent

I downloaded the model from the CMU Perceptual Computing Lab. The model is 200mb, and it was trained on this 20gb dataset. On my lower-power computer, it took 16 seconds to process each frame of video.

This code runs the model:

import cv2
net = cv2.dnn.readNetFromCaffe(protoFile, weightsFile)
inputBlob = cv2.dnn.blobFromImage(source_image, parameters)
output = net.forward()

I have had better results with higher resolution video that has more contrast. These overlays show the accuracy of the model. To create this video, I passed an inputBlob array of shape (1, 3, 368, 368) to the model, which is slightly larger than the size of your body in the video.

These results are head an shoulders better than anything that I have seen. Running these models is quite easy, and would take less than an hour (which includes time to install Ubuntu and OpenCV) and can be done on an older machine. This is a good afternoon project for an intern.

Higher frame rates (120fps+) are required for ball tracking, but greater resolution is required for human pose estimation. I do want to try this out on more video, but I don't even know what to ask for at this point! If you have a video clip for me to process, you can send it to


Subscribe to this forum via RSS
1 article per branch
1 article per post

Green Eggs reports