3D Registration in Dark Environments Using RGB-D Cameras

This paper presents a new approach to align corresponding 3D points obtained by different frames in environments with varied illumination using an RGB-D camera (Microsoft Kinect). Our method switches between the RGB and IR images for feature extraction based on the brightness level of the images. The corresponding visual features are matched using their descriptors and outliers (false matches) are removed using a rank ordered statistics based robust estimation approach. The estimated 3D transformations are finally refined using an Iterative Closest Point (ICP) approach. We show that our method is able to obtain accurate transformation estimation between frames in dark environments (typical office environments with no artificial lighting). We finally present a real-time Visual Odometry (VO) system that concatenates the estimated camera transformations between sequential frames and obtains a global camera pose estimate with respect to a fixed reference frame that outperforms the state-of-the-art methods in both lit and dark environments.