Depth filtering using total variation based video decomposition

Abstract

In vision based applications, depth plays a crucial role from different aspects. From 3D rendering to automation, precision in depth measurement is important for acceptable performances. Though several techniques have been proposed to capture depth map of a scene, the estimation is either erroneous or much expensive for regular usage. Thus, the demand for high accuracy depth measurement is prominent in the field of robotics and computer vision. In this paper, we propose a method to estimate high accuracy depth map from a raw depth map for both static and dynamic scenes. This depth filtering is done by exploiting the spatio-temporal information present in a depth video and by taking the color information of the scene into account.