Hello, I run in a machine learning/AI lab oriented towards medical application. We would like to create our own high-resolution dataset for the purpose of characterizing the movements of patients of all ages in a hospital environment. We can't replicate this adequately in a lab environment using labeled motion capture.
As a second-best option, we would like to acquire multi-view video from 3 cameras with distinct views on the hospital room. Originally, we were planning to do this with conventional cameras, but it seems that a set of the Oak-D cameras (OakD pro wide) might be able to give us even more data for the same number of enclosures placed.
We would love to get both the 1080p@60fps and the depth estimate from the cameras and have IR capability for night-vision and stream the results to host computer(s) for post-acquisition analysis.
I wanted to see if anyone has done this? If so what sort of problems need to be overcome.
Questions that occur to me:
- do I need a separate device to synchronize the cameras
- do the IR laser dot projectors interfere with each other or improve things by giving a denser "dot texture"
- can one host computer handled the I/O stream.
Our hope is that once we have our dataset, we would acquire more OAK cameras to run patient/hospital tuned models.
Many thanks in advance