Deep reinforcement learning for frontal view person shooting using drones

Abstract

Unmanned Aerial Vehicles (UAVs), also known as drones, are increasingly used for a wide variety of novel tasks, including drone-based cinematography. However, flying drones in such setting requires the coordination of several people, increasing the cost of using drones for aerial cinematography and limiting the shooting flexibility by putting a significant cognitive load on the director and drone/camera operators. To overcome some of these limitation, this paper proposes a deep reinforcement learning (RL) method for performing autonomous frontal view shooting. To this end, a realistic simulation environment is developed, which ensures that the learned agent can be directly deployed on a drone. Then, a deep RL algorithm, tailored to the needs of the specific application, is derived building upon the well known deep Q-learning approach. The effectiveness of the proposed technique is experimentally demonstrated using several quantitative and qualitative experiments.