Monocular vision navigation system for UAV autonomous mission: a real-time window-based obstacle avoidance approach

The need for autonomous unmanned aerial vehicles (UAVs) is rapidly increasing for various industrial applications today. However, the realization of UAV autonomy requires a careful mix of Artificial Intelligence (AI) and computer vision (CV) techniques to be implemented to address end-user challenges and satisfy field requirements. This study aims to develop an intelligent system for UAV to achieve autonomy with standard low-cost hardware components. We have infused state-of-the-art AI and CV technologies into our developed system architecture to improve autonomous navigational capabilities of UAVs. The proposed architecture has been implemented and tested on the DJI Tello UAV equipped only with a monocular camera on-board and an inertial measurement unit (IMU). We focused on solving multiple practical technical problems that have arisen during this study due to the hardware limitations of the UAV. The study investigated methods for representing obstacles within an environmental map, which are crucial for effective avoidance and path planning, such as their edges, width, and shape, to ensure accurate detection and navigation. The experiments were carried out in an environment that included dynamic obstacles that were not considered in the map. The experimental results showed outstanding performance in terms of successful autonomous navigation for a wide range of light intensity. These results show that the proposed approach has a high potential to achieve autonomous UAV operation in the field with low hardware and energy requirements.