Tracking Micro Aerial Vehicles with High Maneuverability and Sudden Rotation in Complex Indoor Environments Using Deep Neural Networks in Three Dimensions based on the Received Signal Strength

Abstract

Today, with the rapid growth of indoor location-based services, location methods in three-dimensional indoor environments have attracted researchers. Most of the proposed methods, assuming the limited maneuverability of the target node, have high accuracy, which requires to expensive implementation of devices with high processing capability. In practice, due to the existence of various fixed and moving obstacles in a complex indoor environment, the movement of the target node is accompanied by a sudden rotation. Some of these obstacles include the use of micro aerial vehicles in complex indoor environments that are used significantly today and many of the services provided, such as data protection and monitoring of specific locations, require tracking location. In this study, a deep neural network-based method for tracking micro aerial vehicles in a complex indoor environment using a wireless sensor network based on the Received Signal Strength method without requiring to additional hardware is introduced. Deep neural network training use Back-Propagation (BP) algorithms based on gradient descent and the Unscented Kalman Filter (UKF). The simulation results show that the deep neural network training with the UKF gives better stability than the back-propagation method based on gradient descent. Also, the neural network trained by the UKF has high accuracy in estimating locations.