We present a complete system for micro-aerial vehicle autonomous navigation from vision-based sensing. We focus specifically on mapping using only on-board sensing and processing, and how this map information is best exploited for planning, especially when using narrow field of view sensors in very cluttered environments. In addition, details about other necessary parts of the system and special considerations are presented. We compare multiple global planning and path smoothing methods on real maps made in realistic search and rescue and industrial inspection scenarios.