Improve hand detection aggregation

[wolterlw]

Currently hand detections are not aggregated, instead the detection with the largest bounding box size is selected. This often leads to incorrect bounding box rotations, which impedes landmark detection.
Accuracy can be approved by implementing non-max suppression or a variation of it.

[ausaf-a]

I'm interested in attempting, is this still active?

[wolterlw]

Sure, you can give it a go.