Efficient visual odometry and mapping for Unmanned Aerial Vehicle using ARM-based stereo vision pre-processing system

Visual odometry and mapping methods can provide accurate navigation and comprehensive environment (obstacle) information for autonomous flights of Unmanned Aerial Vehicle (UAV) in GPS-denied cluttered environments. This work presents a new light small-scale low-cost ARM-based stereo vision pre-processing system, which not only is used as onboard sensor to continuously estimate 6-DOF UAV pose, but also as onboard assistant computer to pre-process visual information, thereby saving more computational capability for the onboard host computer of the UAV to conduct other tasks. The visual odometry is done by one plugin specifically developed for this new system with a fixed baseline (12cm). In addition, the pre-processed infromation from this new system are sent via a Gigabit Ethernet cable to the onboard host computer of UAV for real-time environment reconstruction and obstacle detection with a octree-based 3D occupancy grid mapping approach, i.e. OctoMap. The visual algorithm is evaluated with the stereo video datasets from EuRoC Challenge III in terms of efficiency, accuracy and robustness. Finally, the new system is mounted and tested on a real quadrotor UAV to carry out the visual odometry and mapping task.