Automatic video to stereoscopic video conversion

Abstract

In this paper a method to convert a monoscopic video movie to a stereoscopic video movie is presented. This method is based on passively acquired video images using a single camera and may be applied in PC based real time system. Current methods for generating single camera stereoscopic videos and ad-hoc standards are based on creating a depth map. The depth map calculation is based on structure from motion methods. In order to work properly the depth map should be very dense and accurate. Otherwise, local deformations may occur. Our proposed method is based on calculating a planar transformation between images in the sequence and relies on the human capability to sense the residual parallax. Therefore it does not depend on calculation of the depth map. This advantage is of great importance when deformation of even small objects is forbidden as in reconnaissance and medical systems. The proposed method generates stereoscopic image pairs. Each pair consists of the original image and a transformed image. The transformed image is generated from another image of the original sequence. That image is selected by the algorithm such that a considerable parallax is developed between the two original frames. The chosen image is then warped using the planar transformation. Our algorithm assumes a good planar transformation approximation between the images. Thus, residual disparities are related to the distance from the average plane. As such, zooming and rotation are eliminated after the warping process, and the eyes integrate the images to 3D scene based on the disparity. The algorithm has been tested and is demonstrated on aerial scenes as well as terrestrial using a hand held camera.