3D registration of point clouds from time-of-flight cameras

A novel approach for aligning two 3D point clouds acquired by time-of-flight (TOF) cameras is presented. Given that the density of 3D data of these devices is not high, the alignment will be carried out using descriptors based on a direct encoding of the environment of a set of previously selected points-of-interest. This will allow the design of a similarity measure that behaves correctly when both the density of the point clouds and the overlap between them are low. Furthermore, this combination of descriptor and similarity measure facilitates the alignment of point clouds from devices with different accuracy and resolution (such as a laser scanner and a TOF camera).