Low memory footprint mode using memory management only for visual features

[matlabbe]

See original discussion here: http://official-rtab-map-forum.206.s1.nabble.com/Memory-management-in-localization-mode-td9886.html

The idea is to have in RAM the global occupancy grid (don't keep local grids), the global pose-graph (including nodes from LTM and WM) and visual dictionary based only from nodes in WM.

The why:

• On very large maps, the visual dictionary can take GBs of RAM. If the robot is globally localized in the map, we don't need to keep in RAM all visual words of all nodes in the map. The visual dictionary will still be updated with close nodes to current location that were in LTM and brought back in WM.

[alexk1976]

Other possible option - to load/build dynamically Dictionary of the relevant area only. Assuming we know (based on navigation task) what part of the map robot is navigating- theoretically it could be possible to build Dictionary of relevant part only of the map and not whole/huge map.