Recognition of 3-D objects from computationally reconstructed integral images using 3-D reference image

Abstract

Recently, many research works are actively being done on three-dimensional (3-D) recognition using correlation method throughout the world. To recognize the 3-D object from the target image, the reference image is needed. That is, target image is captured using lenslet array as a form of elemental image array (EIA) by integral imaging method. Then, the EIA is reconstructed at each depth plane as a form of plane object image (POI) using computational integral imaging reconstruction method. Then, the correlation process is performed at all POIs with reference image. In general, twodimensional image that is the scene from the front view of the object is used as a reference image. Thus if target has big depth range, the reconstructed POIs are not clear at all depth range because some part is focused whereas the other ones are blurred. Therefore, there are no POIs having clear 3-D object. In other words, it is not able to find the accurate location of 3-D object if 2-D image is used as a reference image. Accordingly, in this paper a new method to find the accurate location of 3-D object by using 3-D reference images that is captured by integral imaging method is proposed. That is, the correlation process is performed the reconstructed POIs of target image and reconstructed POIs of reference image at all depth range. To confirm the proposed method, 'car' is used as a 3-D object. From this simulation results confirmed the feasibility of the proposed method to extract the location of 3-D object.