Muhammad Shoaib1 , Nasir Sayed
n recent years, integrating artificial intelligence and unmanned aerial vehicles (UAVs) has become a hot topic of study, especially where UAVs must conduct complex tasks that cannot be completed quickly under human control. Drones often use several sensors to gather full details about conditions, such as a top-down camera or LiDAR sensors, and the main processor measures all of the drone’s trajectories. This paper proposes tracking a detected target that employs a monocular on-board camera and a reinforcement learning model. This system is more cost-effective and adaptable to the atmosphere using various sensors and precalculated trajectories than previous approaches. Our model added encompassing box details to the drive network picture input by extending the previous Deep Double Q network with the Duel Architecture Model (D3QN), modifying an action table and incentive feature, enabling 3-dimensional gestures and object recognition combined with MobileNet’s support. The simulations are carried out in various simulation settings, each with its level of difficulty and sophistication. The “Airsim” application, a Microsoft-supported quadrotor simulation API, is used for research. The findings reveal that using a convergence-based exploration algorithm, the model approaches the observed object, a human figure, without reaching any barriers along the way and is moved faster.
English 
Spanish 
Chinese 
Russian 
German 
French 
Japanese 
Portuguese 
Hindi