Separate capture from depth calculation

[ngeorgis]

It would be nice to have the ability to capture raw data from sensors and do not run any processing during capture. This would enable lower resource CPU/GPU or no-GPU systems to be used as capture nodes.

In the current implementation my understanding is capture of raw depth is not possible. Instead there is a requirement for a fast-enough GPU to be used for depth engine calculations running on the capture system. In some workflows it is better to capture in real-time, and run depth engine calculation later, not in real-time. We could also choose to run own custom depth engine in real-time and we do not want the GPU to be busy as currently is the case. I could not find any way to implement this two-phase Capture/Process approach in the current SDK but I may have missed this feature.