Autonomous 3D mapping and surveillance of mines with MAVs

Abstract

The mapping of mines, both operational and abandoned, is a long, di cult and occasionally dangerous task especially in the latter case. Recent developments in active and passive consumer grade sensors, as well as quadcopter drones present the opportunity to automate these challenging tasks providing cost and safety bene ts. The goal of this research is to develop an autonomous vision-based mapping system that employs quadrotor drones to explore and map sections of mine tunnels. The system is equipped with inexpensive, structured light, depth cameras in place of traditional laser scanners, making the quadrotor setup more viable to produce in bulk. A modi ed version of Microsoft's Kinect Fusion algorithm is used to construct 3D point clouds in real-time as the agents traverse the scene. Finally, the generated and merged point clouds from the system are compared with those produced by current Lidar scanners.