Raspberry pi, kinect, Mac, ROS, halp?


#1

I have a student (more of a client at this point) who would very much like to get some motion capture data of themselves. We tabled the project this summer after a few hardware configurations didn’t work out, and it’s turning into such a rabbit hole that I feel it’s worth to ask.

I have a few questions, but what I really want is someone I can sit down with who maybe has some of this hardware (my client is in California and I’m working remotely) and can explain some of the issues. Since this is for work, I am willing to compensate you for your time.

My client does not have reliable access to a PC, only a Mac and a Raspberry Pi 3 that I recently advised them to get. I’m paid by their parents, but the client pays for any hardware themselves, and I understand a new PC is out of the question, hence my recommending the RasPi 3. I’ve recently learned that the Kinect v2 uses USB 3 and the RasPi 3 only has USB 2.0, in which case I probably advised them to get incompatible hardware.

My first question is, am I correct that I messed up? My second question is, how might I best minimize the damage?

Again, I am very much willing to pay fairly for your time if you can spare any to meet with me, especially since this is highly skilled and specialized help I’m seeking. Thank you.


#2

Looks like some people are using neural networks to do this with a single regular webcam. In the AI space, this goes by the name “Human Pose Estimation”. Google search ‘neural network “human pose estimation” medium’.

Here is one which you can run in a web browser.
https://storage.googleapis.com/tfjs-models/demos/posenet/camera.html
Corresponding article:
"Real-time Human Pose Estimation in the Browser with TensorFlow.js” by TensorFlow https://link.medium.com/Ad1eHcZkoT

Another version you can run in Python:



#3

I don’t even. Whoa.

I’m still looking for raspi help, but this… interesting.


#4

it’s difficult to say without more detail of your project. I’ve worked on a similar project, sans the kinect. Send me a message or email at [email protected] and we can discuss


#5

Agreed - some more info would be helpful. Do you just need to get the body points? Display the body frame on a screen?


#6

The points. Visualization is something we can do any old way, that won’t be an issue, we just need to know where the person’s joints are in time, every fraction of a second.