Enhanced Model-Free Deep-Q Network based PTZ Camera Control Method

Recently, the algorithm based on reinforcement learning for controlling PTZ(Pan/Tilt/Zoom) of a camera has been significantly increasing. To improve the accuracy of video analysis in a video surveillance environment, it is essential to get good video sources from a camera such as an appropriate object size or an object not covered by the other object. In this paper, we propose a model-free reinforcement learning based PTZ control method to support improving the accuracy of video analysis. The proposed method fine-tunes the DQN (Deep-Q Network) for smoothly controlling PTZ camera. The simulation results show that the proposed method can automatically and seamlessly manage the PTZ camera according to moving objects.

[1]  Guangwen Yang,et al.  Episodic Memory Deep Q-Networks , 2018, IJCAI.

[2]  Peggy Seriès,et al.  Reward-Based Learning, Model-Based and Model-Free , 2014, Encyclopedia of Computational Neuroscience.

[3]  Shane Legg,et al.  Human-level control through deep reinforcement learning , 2015, Nature.

[4]  Dongchil Kim,et al.  Automatic PTZ Camera Control Based on Deep-Q Network in Video Surveillance System , 2019, 2019 International Conference on Electronics, Information, and Communication (ICEIC).

[5]  Anil A. Bharath,et al.  Deep Reinforcement Learning: A Brief Survey , 2017, IEEE Signal Processing Magazine.

[6]  P. Dayan,et al.  Reinforcement learning: The Good, The Bad and The Ugly , 2008, Current Opinion in Neurobiology.