DQN and TensorFlow Agent-Based Visual Object Tracking Algorithm using Deep Reinforcement Learning

Object tracking models are multi-functional algorithms that are applied in many fields of uncertain environment tracking situations. In addition, object tracking in physical environments and conditions is a much more difficult task to obtain accurate results. However, the process can be tested by evaluating and verifying the performance of the model in various simulation conditions by experimenting with a virtual simulator. In this paper, we propose a visual object tracking algorithm based on DQN and TensorFlow agents using deep reinforcement learning. TensorFlow agents are trained in the Blocks environment to adapt to the environment and existing objects in the simulation environment, and further tests and evaluations are performed on the accuracy and speed of tracking. The DQN algorithm uses a deep reinforcement learning model to investigate the environment in a virtual simulation environment using sequential photos of the virtual environment simulation model. The proposed DQN and Tensor Flow agent-based process and the deep reinforcement learning-based object tracker are tested and compared with existing methods, and the results show that the proposed DQN and TensorFlow agentbased process is superior in terms of stability, speed, and numerical performance.