Confidence indicators based pose estimation for high-quality 3D reconstruction using depth image

Abstract

Real-time 3D reconstruction using depth information has become a critical part of many emerging applications, like augmented reality, autonomous robot navigation and teleoperation to hazardous locations. This paper presents two novel methods to improve on the 3D reconstruction quality of static scenes using depth images. First, a new anisotropic depth filtering along with confidence indicator for every filtered depth value is proposed. Second, a modified pose estimation method, which incorporates the uncertainty or error characteristics of the depth measured as part of the Iterative Closet Point (ICP) algorithm is proposed. Both these methods are combined together along with a pre-existing confidence indicator based 3D reconstruction method, to get a robust and real-time 3D reconstruction framework. Simulations results show that this combined algorithm gives 59.60% quantitative improvement in pose estimation and achieves a significant quality improvement in 3D reconstruction compared to the baseline algorithm.