3D indoor reconstruction using Kinect sensor with locality constraint

In this paper, an indoor 3D construction is proposed based on RGB-D measurement. It is intentionally designed to solve the traditional issues, such as cloud registration inaccuracy, large computational time. Firstly, potential candidates are extracted by Harris detector, and the SURF method is used to generate the feature descriptors. Afterwards, the correct functional match is selected by RGB and depth measurements with neighbouring constraint. Lastly, 3D clouds are formed through graphical optimisation. In the experiment, the RGB-D sensor is rigidly fixed on the mobile platform to reconstruct the indoor 3D scene, which shows comparable performance in terms of computational time and accuracy.