Depth Recovery For Binocular Kinect RGB-D Cameras

These years,stereoscopic displays and three-dimensional applications are more and more popular in the consumer market.And in these applications,high quality depth information is of the essence.However,the depth images captured by most main-stream method are incomplete that influences the visual effect and development of 3D application.Therefore,how to acquire high-quality depth map has been being a signif-icant task.This paper builds a binocular acquisition system using Kinect v2 cameras,and proposes a novel depth recovery framework for this binocular system to obtain high quality depth map.The results and innovations of this thesis are summarized as follows:1.This paper proposed a novel binocular acquisition system with two Kinect v2 cameras for capturing the depth and color information of the scene simultaneously.This system is simple in structure and scalable that can be extended to a multi-view acquisition system with three or more cameras.In addition,the error characteristics of this system is carried out for providing prior information for the follow-up work.2.This paper proposed a progressive depth curve estimation for generating high-resolution depth curve map.This method utilizes the depth edge of low-resolution(LR)depth map,registered depth map and color edge to estimate high-resolution(HR)depth curve via Moving Least Square(MLS).And this estimated depth curve information can be helpful to avoid artifacts around boundaries in depth refinement.3.This paper proposed a progressive depth fusion method for binocular view.The edge-guided depth fusion strategy is proposed via error compensation based on sys-tematic characteristics.This strategy fully uses the advantage of binocular system over single view,achieves the depth compensation of one view to another.4.This paper proposed a progressive depth refinement method.This method mod-ified the order of inpainting to make depth values propagate from boundary of holes to the inside and stop when meeting depth curve.Meanwhile,depth contour and col-or edge are joint used to guarantee the accuracy of depth propagation.Experimental results show that the proposed method can achieve excellent restoration performance even in the case of large depth missing.