Optimizing Indoor Localization and Tracking: An Energy-Efficient Approach Using Received Signal Strength and Mixstyle Neural Networks With Implicit Unscented Particle Filtering

Abstract

In indoor environments, the unpredictable noise in received signal strength indicator (RSSI) measurements causes very high estimation errors for target localization. Nowadays, RSSI-based localization systems are widely used to deal with higher noise levels in RSSI measurements and to assure more accuracy in target localization. In this paper, Optimizing Indoor Localization and Tracking: An Energy-Efficient Approach Using Received Signal Strength and Mixstyle Neural Networks with Implicit Unscented Particle Filtering (OILT-MNN-IUPF) is proposed. The proposed method consists of two range-free target localization schemes in wireless sensor networks (WSN) for an indoor setup: (i) mixstyle neural network (MNN) used for regression tasks and (ii) fusion of MNN and the implicit unscented particle filter (IUPF). The fusion-based model is named the MNN + IUPF approach. There is no need to compute distances using field measurements for the proposed localization solutions, here three RSSI measurements to trace the mobile target. Also, this paper discusses the energy consumption related to the typical trilateration and MNN-based target localization. With the proposed MNN-based schemes, linear, sigmoid, RBF, and polynomial are the four kernel functions estimated on the accuracy of target localization. The proposed OILT-MNN-IUPF model achieves 25.05%, 20.17%, and 23.19% lower average localization error and 23.11%, 20.11%, and 24.09% less root mean square error compared with existing models.