Virtual Simulation based Visual Object Tracking via Deep Reinforcement Learning

Abstract

The current research field of object tracking has become noticeably popular among researchers where AI techniques take place with high-level accuracy. An algorithm with multifunctional abilities had proposed in different proposals in recent years. We proposed a tracking technique integrated with a virtual reality simulator – the AirSim (Areal Informatics and Robotics Simulation) City Environ model using one of the DRL models to control with a drone agent to examine a realistic environment. Additionally, the suggested method had tested via the two public: VisDrone2019 and OTB-100 datasets to compare with conventional strategies to show better performance among recent works.