How to Combine Color and Shape Information for 3D Object Recognition: Kernels do the Trick