Active Stereo Method for 3D Endoscopes using Deep-layer GCN and Graph Representation with Proximity Information

3D endoscope system with AR display superimposing dense and wide-angle-of-view 3D points obtained by using micro pattern projector

“…3(a), the 3D endoscope system acquire depth information with active stereo system which is composed of the endoscopic camera and a small-scaled pattern projector that can be inserted through the instrument channel (Fig. 3) similar to Furukawa et al [18].…”

“…Although the proposed system is similar to Furukawa et al [18], the proposed pattern projector has been improved in two ways. First, compared to [18], the projector of the proposed system has wider angle of projection. Thus, the maximum area of the depth measurement in the endoscopic camera largely increases.…”

“…Second, the projected pattern has a new codification strategy. In [18], each grid points had codes from three symbols using line gaps for only vertical direction. In the pattern of the proposed system, each grid points had codes from five symbols by using gaps for both vertical and horizontal directions.…”

“…The graph is then processed by a GCN to produce node-wise correspondence prediction. The GCN is trained using dataset generated from CG-synthesized images similarly with [18].…”

“…We measured objects such as medical phantoms using the proposed method, and compared the reconstruction results with a measurement result captured with a previous method [18]. The results are shown in Fig.…”