Depth Filtering in 3D Reconstruction of Indoor Scenes Based on Kinect

Recently due to the appearance of low-cost 3D camera, such as Microsoft Kinect, 3D reconstruction plays an increasingly important role in computer vision fields, as well as image processing and pattern recognition. But owing to the physical shadow and the underlying generating mechanisms, the range map from Kinect contains a large number of noises. In this work, a new 3D reconstruction workflow aiming at the above problem is presented, including hole-filling, coordinate transformation from 2D to 3D, and 3D depth de-noising algorithm. A 3D indoor scene reconstruction example is presented to demonstrate effectiveness of the proposed workflow. The experiments results illustrate that the provided method can reconstruct an indoor scene with less memory cost but good performance.