《Fast and robust 3D feature extraction from sparse point clouds》，看一下前端是否可以用到。前端学习特征提取，也需要加上tracking的功能。这样可以加速。
http://jacoposerafin.com/wp-content/uploads/serafin16iros.pdf
https://github.com/HiltyBosch/PCLPlaneAndColumnDetect
A comprehensive review of 3D point cloud descriptors
Efficient Surfel-Based SLAM using 3D Laser Range Data in Urban Environments
http://jbehley.github.io/projects/surfel_mapping/
Comparison of the Selected State-Of-The-Art 3D Indoor Scanning and Point Cloud Generation Methods
segmap
ETH出品的
Learning-to-Localize-Using-a-LiDAR-Intensity-Map.pdf
http://n.ethz.ch/~cesarc/files/ICRA2017_rdube.pdf
segmatch
