Coverage maximization using deep reinforcement learning for mobile video camera networks

Abstract

Mobile wireless sensor networks (M-WSN) represent networks of movable sensor nodes which observe data from the environment. Maximizing the observable area of the sensor nodes, coverage maximization, is an important issue of the M-WSN. Several works have dealt with the issue in the past. They assumed the sensing range of each sensor is circular and found the position of each sensor node maximizing the sensing range. Circular sensing range is a reasonable assumption for most types of sensors. However, for visual sensors like video camera, which have a specific sensing range, field of view, it is not reasonable to assume that the camera could observe circular range. And for visual sensors we should consider not only the effect of obstacles on the position of the sensors but also on the sensing range. In this paper, we deal with the coverage maximization problem for specific sensor, video camera, assuming the fan shape of cover range. The goal of this paper is to maximize coverage of video camera networks with avoiding obstacles in dynamic position. We use deep reinforcement learning which is proper to apply for environment with dynamic condition. Simulation results show that our method attained sufficient performance.