Towards using covariance matrix pyramids as salient point descriptors in 3D point clouds.

In this work, a novel salient point descriptor for 3D point clouds, called Covariance Matrix Pyramids (CMPs), is presented. With CMPs it is possible to compare unstructured and unequal numbers of points which is an important characteristic when working with point clouds. Corresponding points from different scans are matched in a pyramidal approach combined with Particle Swarm Optimization. The flexibility of CMPs is demonstrated on the basis of several databases with objects, such as 3D faces, 3D apples, 3D kitchen scenes, 3D human–machine interaction gesture sequences, and 3D buildings all recorded with different 3D sensors. Quantitative results are given and compared with other state-of-the-art descriptors, whereby CMPs show promising performance.