Depth-Based `SmartMotion` Feature

[Luxonis-Brandon]

Start with the why:

The pixel-motion of https://github.com/luxonis/depthai/issues/132 , although plenty useful in indoor applications/etc. may not give the intended behavior (detecting things that are physically moving in the field of view of the system) as shadows, changes in lighting, etc. will cause pixel-motion even though there may be physical motion in that region.

Move to the how:

Implement the same flow as in https://github.com/luxonis/depthai/issues/132, but where the motion is depth-map based instead of pixel-motion based.

Experimentation will likely be necessary to figure out which parameters need to be settable from the host side. Potential examples include:

1. What depth confidences to ignore motion from (i.e. below a threshold)
2. Minimum motion size (e.g. how many pixels)

Move to the what:

Use depth-based motion detection to inform the cropped size(s) to pass to an object detector which covers only that cropped area, then feed that result to an object tracker which tracks on full-frame to cover cases where the object stops moving.