I have implemented a better way to incorporate several detections using detect_all_faces.

When a single face should be detected, it merges all overlapping detected bounding boxes that are over threshold into one single bounding box. In the old way, when multiple faces should be detected, for overlapping detected bounding boxes, only the one with the highest detection value was returned.

In the new version, also for multiple faces, overlapping bounding boxes are merged into one, similar to the single face case.

@andre.anjos, @tiago.pereira, @amohammadi : Could you please check this?