Autonomous Drone Navigation Using Monocular Camera and Light Weight Embedded System

Autonomous robots are machines that can act without any human interference. When certain sensors and decision making algorithms are added to the control unit of a drone, the aerial vehicle is said to be autonomous. Such vehicles are capable of avoiding obstacles and correcting their local paths while staying on planned global paths. Autonomy in drones is usually achieved using many types of sensors like depth sensor, stereo camera or lidar, and intensive SLAM algorithms that require powerful processors. Though the existing methods work well, the scalability of such products is questionable as the economic and resource availability factors come into play. This paper proposes a method of achieving autonomous navigation using light weight embedded system and affordable monocular cameras by combining features of image processing and resource sharing. The proposed architecture makes use of monocular cues and midas depth estimation model to achieve obstacle avoidance and can run on any processor with basic features such as serial communication, wifi and camera ports.