Vision based Autonomous QuadCopter Navigation through Narrow Gaps using Visual Servoing and Monocular SLAM

In this paper we show a vision based autonomous quad-copter navigation through a narrow gap or window. The goal is to provide an effective solution to a problem of urban UAV search and rescue mission scenario, such as, fire detection and quenching situation in a high rise building. Such scenario may demand a GPS denied navigation and requires extensive use of on-board vision sensors and processors. In this paper, using on-board vision sensor based localization and visual servoing, we show a method to demonstrate a UAV autonomous navigation to enter a building through a narrow window. The entire mission scenario is divided into two phases, where the first phase comprises an effective visual servoing based approach trajectory generation and execution. In the terminal phase, the vision sensors may fail to locate the desired target features due to the constraints in view angle geometry. To compensate that a non-linear guidance law based trajectory following method is used which generates a monocular SLAM based phantom target locus and provides a consistent homing control. We have experimentally demonstrated and evaluated the method in a real world scenario with substantial success rate.