비주얼 헐 보호막을 활용한 가려짐에 강인한 손안 물체 스캐닝

Abstract

We present an occlusion-robust in-hand object scanning system using visual hull shield. Our system basically uses depth-based volumetric method for 3D object reconstruction. In in-hand object scanning, due to a small object size, most of previous depth-based volumetric methods are sensitive to outlier depths caused by occlusions, e.g. hand occlusion. To overcome this limitation, we generate visual hull shield using object silhouettes and remove outlier depths that cause the drift of the object pose estimation. Experimental results show that the proposed method is more robust to occlusions than previous 3D object scanning algorithms.