We show how a local graph-based representation can be used for structuring 3D point clouds provided by multiple LiDAR acquisitions. Those graphs are defined from the depth maps generated by the sensors, and connected together to get a single and global representation of the scanned site. Experimental results show that this structure is particularly suitable for processing gigantic points clouds, even if they largely exceed the memory capacity of the machine on which those processings are performed, whatever the number of points and of acquisitions.