Monocular Vision-based Obstacle Avoidance Trajectory Planning for Unmanned Aerial Vehicles

Monocular vision has become a promising sensing method for Unmanned Aerial Vehicle (UAV) Sense and Avoid (SAA), with the advantages of low cost and small size. However, obstacle perception capability of monocular vision is limited due to the restriction of optical imaging properties, which have significant influence on avoidance trajectory planning. By considering the characteristics of monocular vision-based optical measurement, a receding horizon-based collision avoidance trajectory planning algorithm is proposed for eliminating the hazard of both static and dynamic obstacles in this paper. This paper aims at dealing with two core problems: 1) how to localize obstacle via monocular vision; 2) how to generate a collision free trajectory with partially observed obstacle information obtained by monocular vision. To solve the first problem, monocular vision-based optical measurement for obstacle estimation is firstly analyzed and orthogonal iteration-based localization is further adopted. To solve the second problem, with the constraints containing obstacle avoidance and UAV aerodynamics, a receding horizon-based collisionfree trajectory planning method is proposed. Simulation results demonstrate the algorithm proposed in this paper increases the safety level of UAV and is capable of avoiding both static and dynamic obstacles.