Plenoptic Camera Viewpoints Rotation to Improve 3D Face Reconstruction

Recent advances in plenoptic technology offer new opportunities in the development of video surveillance systems. In this work we assess rotating large-baseline viewpoint cameras for accurate face-reconstruction at large depths. A first study with a 3D plane textured with a smooth gradient shows improvement of reconstruction accuracy by rectifying viewpoints rotation. A second study approaches a surveillance setting of a room with a pan-tilt-zoom camera-array. A human face is synthesized in the middle of the room, allowing to obtain ground truth depth. Results show the array of cameras provides a reconstruction accuracy similar to a near-target plenoptic camera despite of the expected larger face-self-occlusions.