| （英） |
Virtual transparency of visual obstructions can be achieved by light field transmission based on a combination of acquisition and reproduction for 3D visual media. In order to develop an inexpensive system toward virtual transparency, we have previously proposed real-time reconstruction of entire 4D light fields suitable for glassless multi-view displays from point cloud data acquired by a single RGB-D camera. Actually, by applying appropriate geometrical transformation to the acquired RGB-D data, we obtain a simple 3D array while effectively reducing their redundancy. Moreover, each sub-aperture image on 4D light fields can be reconstructed by just structured analysis searching parallel lines within the 3D array, where its computational cost is significantly suppressed. In this paper, we improve our proposed method without additional computational cost for robustly reproducing 4D light fields on a glassless multi-view display with wider viewing angles. Specifically, each sub-aperture image is corrected by changing its origin and scale to suppress crosstalk for a certain viewing distance while keeping wider viewing angles on our multi-view displays. Fortunately, since such correction can be easily integrated with pre-processing to prepare lookup tables for searching the 3D array mentioned above, our improved virtual transparency is still expected to work in real time. |