Kinect sensor′s depth image enhancement based on gaussian mixture model

Depth image acquisition is currently a significant research topic in the field of three-dimensional video processing.The Microsoft′s Kinect sensor can generate a real-time dense depth map.However,there are usually numerous hole artifacts in the depth map which is instable and seriously affects its applications.To solve this problem,in this paper,we present a real-time refinement algorithm for kinect depth map based on Gaussian mixture model.First,we take color image and depth image to be aligned and cropped,then make the foreground and background image separated by GMM,fill the background holes using the background image,apply color matching algorithm for foreground holes.Finally,median filter is used to remove noise.Experimental results show that this method is very effective,for both complex foreground objects and objects with large areas of missing values.Big 〖JP2〗and small holes of depth image are filled, edges are retained,especially good results are achieved in the shade object reconstruction