Low-cost camera based sense and avoid in unmanned aerial vehicles: Sensing and control methods

Over the past two decades, unmanned aerial vehicles (UAVs) have attracted more and more attention because of their increasing applications in the military and civil sectors. Collision avoidance is a core problem in developing UAVs. In this paper, a low-cost camera based sense and avoid control scheme is proposed. First, for the unmanned aerial vehicle (UAV) sense and avoid problem, we propose and demonstrate a low cost solution with only one monocular, wide field of view camera. We adapted a deep learning based algorithm and made it lightweight enough to run on-board the UAV. We created a synthetic dataset to increase the size of the training data set. Second, we developed a filter to estimate the 3D position and velocity of the moving obstacle. Third, based on the estimated information, we adapted the velocity obstacle approach to work in 3D. Finally, we implemented the algorithms in the software framework of our UAV testbed and conducted flight tests, demonstrating the effectiveness of the solution.

Automated summary

This paper proposes a sense and avoid control scheme based on a low-cost camera, using deep learning algorithm and velocity obstacle approach to achieve UAV sense and avoid, demonstrating the effectiveness of the solution through flight tests.