The variance estimation and enhanced 3D sensing of heavily occluded objects using synthetic aperture integral imaging (SAII)

Abstract

We present enhanced 3D object reconstruction of heavy occluded object and enhanced 3D object recognition using variance estimation with synthetic aperture integral imaging (SAII). SAII is a technique that acquires the elemental images moving imaging sensor of camera without the micro lens array which was used to get elemental images in conventional II system. We used volumetric II among various computational II. In volumetric reconstruction, the focused areas of the reconstructed image are varied with the distance of the reconstruction plane. A partially occluded object which is focused can be reconstructed clearly by making the occluding object blur using the lens. However, to reconstruct an unobstructed object using volumetric II in heavy occluding object, other methods are additionally required because occluded object is reconstructed with heavy blurred distribution of occluding objects. The proposed method is to get 3D unobstructed object using the value of variance of superposed pixels of elemental images for heavy occluding object. If the variance of superposed pixels is calculated, because the variance value of pixels in the only blurred distribution are small comparing with the value including the pixels of occluded object, the heavy occluded object image is reconstructed remedying blurred distribution due to the occluding objects. Calculating this variance, quality of reconstructed image is enhanced. We applied to 3D recognition with the enhanced visualization, and it is shown that the peak value of correlation with enhanced reconstruction using the proposed method is superior to the 3D recognition without enhanced 3D object reconstruction.