Lesson From a Gig using a ZED 2i Camera
HTML-код
- Опубликовано: 4 окт 2024
- Get access to 200+ hours of TouchDesigner video training, a private Facebook group where Elburz Sorkhabi and Matthew Ragan answer all your questions, and twice-monthly group coaching/mastermind calls here: iihq.tv/Trial
For my latest project, I prepared a bunch of visuals and programmed some interactivity as well. Using the ZED 2i camera for the first time without doing proper research was a mistake, and ended in some failures, because the interactive part didn't work when I tested the project just before the show.
In this video, I'll share the lessons I learned from this project, so you can avoid making the same mistakes that I did.
Follow Marco on social media:
Instagram: @reflekkt_
RUclips: @reflekkt_net
This is so great of you not to only share the good moments, but let us in on the failure. And get up again. Stronger.
I like to see an in deep tutorial about your proyect, special things like the tops you use to control more than 2 output windows, how to blend proyectors and the hardware you use to control too many outputs.
The projector blending happens on a media server. I "just" deliver a video in full resolution covering all 4 walls (the video on the bottom at 3:02) and send it via NDI to the server. There are 2 videos about the project on my YT channel, there I talk more about the pixelmap etc.
Marco, thanks for sharing your experience with the camera! There's not a lot of material like this on the web. You often have to test theories through personal experience. Like you said in the video, interactive design requires prototyping. This is where I absolutely agree. There can be a lot of surprises without tests. Personally, I would like to confirm that the conditions of this camera are really not very suited for the current setup. To reduce the tracking error it would be better to place two cameras along the long wall instead of one short one. And to give additional background illumination of people (for example, with the 20-pixel frame of the projection), so that the camera did not lose people during the dark content. But much better here would work 2d lidar or IR camera. I really like that there is Zed2i, as it can completely close the interactive on the street installations. IR cameras on the street do not work as well as I would like. It's a pity that there is no single solution for all interatvic design and for each setup you have to choose your own device. But maybe neural networks and new generations of sensors will come to our assistance?
That's some good advice, thank you. For an upcoming project I'm considering to use a ZED 2i or Realsense camera (it has shorter distance tho), and this time I have some more time to properly test it before :D Will try the background illumination trick as well, maybe that's enough for the ZED to keep the tracking kind of stable.
Regarding upcoming sensors, I'm sure there will be some AI solution for this soon! Even tho the ZED also works with some crazy algorythms already...
Thanks for uploading this practical case. It's really useful to see this kind of things and how to avoid mistakes. We've all been there at some point.
Please, keep uploading this type of videos!
We appreciate the feedback! I've shared it with the team. Glad you found it helpful :)
oh brother. I felt that one 😂 everybody’s been in a situation like that working with depth cams and other sensors at some point. I know I have. but I would’ve made the same assumption, been looking at this specific camera and totally expected that it would work in the dark! so thank you for making and sharing this mistake - it helped me avoid a similar f*kup 🙏
Happy to hear I'm not the only one expecting this :D
Hey, I'd love to know more about your projection setup here if you have a moment! Can I ask how many projectors you're using, and what you used for the projection fabric?
Is there a best in class depth sensor at the moment or are we in a bit of a limbo?
We've had the exact same issues as Marco. We experimented with Zed's as an alternative to the loss of the Kinect Azure and lost many hours troubleshooting in a projection environment. We also tested many other approaches such as using webcams into Blazepose, Movenet etc. In the end we've cycled back to ToF options and are using the Orbbec Femto Bolt - I have to say that I think this is the best in class sensor now for nearly all situations.