An efficient 3-D environment scanning method

In this paper, we discuss an idea of a system that can capture the 3-D model of a large area using only one single Kinect 3-D range sensor plus a stationary master camera. In operation, the Kinect is placed at different key positions to capture the local 3-D models, while a stationary master camera is situated behind the Kinect to find the current pose of the Kinect range sensor. Traditionally, a large scene can be scanned by moving the Kinect sensor across the whole area. Then the models obtained can be combined using motion capturing and pattern matching methods. However, the accuracy deteriorates when the area is too large or the environment does not provide enough features for registration. In our proposal, we place the Kinect at different key positions to obtain a number of local models. A dual-face checkerboard is placed on the top of the Kinect sensor in a way that the pattern can be seen from both the front and rear sides but not blocking the view of the Kinect. The pose of the board and the Kinect is estimated by a pose estimation algorithm using the images captured by the master camera. Since the embedded RGB-camera in the Kinect cannot see the checkerboard, a method based on a mirror is devised to determine the relative pose between the board and the embedded RGB-camera. Finally, we can combine all the 3-D local models and the pose information obtained to build up the complete global model. Various parts of the idea have been tested. We plan to integrate all parts and build a complete system for building the 3D map of a shopping mall or a museum in the future.