Summary

Many applications have a continuous stream of images available to them, especially when using the Kinect. To use more than one image, it is necessary to establish a relation between them. 2-D image registration techniques can be useful under certain camera motions. However, if the camera moves freely in 3-D, a SLAM architecture is necessary. SLAM maps the environment and estimates the camera poses simultaneously. The Kinect is an excellent tool for SLAM, as it simplifies the algorithm considerably and improves its quality. Once the point cloud obtained covers all necessary areas of the scene with sufficient density, surface reconstruction methods can be used to extract a mesh surface from the point cloud.