Direct shape recovery from photometric stereo with shadows (Technion)
As imaging devices move closer to the objects under observation, it becomes increasingly difficult to reconstruct their entire three-dimensional shape. A limited number of current endoscopes and microscopes can address this problem and most of them are based on stereo vision reconstruction or complexly structured light setups. Stereo vision systems are expensive and highly specialized. In addition, two-view stereos are intrinsically problematic due to the often insufficient textural information obtained by two images to accurately reconstruct the surface. While active illumination systems can generate their own texture, they often require complicated and potentially expensive additional equipment.Our new endoscopic imaging system is based on a single view imaging system using 3 or more simple directional light sources. This simple setup is combined with a new and computationally efficient theoretical model understanding the shape of the object from the captured shaded images taken throughout a sequence of different illumination directions of the simple directional light sources. According to the change in shading of the object with respect to a single camera, shape reconstruction can be performed efficiently. We take advantage of this framework and simplify the process of reconstruction in endoscopic situations by augmenting the imaging device with nearby light sources in calibrated positions and directions. The light sources can be as simple as LED's or SMD lights which are easily modeled by our new theoretical framework.
• Reduces the technological and computational complexity required and involves a potentially low-cost physical augmentation.
• Improves existing technologies that already produce 3D reconstructions by augmenting the output results with more accurate details.
• Reduced cost 3D endoscopy for virtual endoscopy registered to live targets through CT imaging
• Close field imaging such as fingerprint analysis, surface modeling, texture modeling
• Low power, low complexity augmentation of existing in-vivo systems
MAE-1513
Israel