Localização em ambientes semiabertos com redução de esforço por super-resolução e fusão de sinais Wi-Fi e FM baseada na distribuição Nakagami-m

Abstract

Localization in indoor and semi-open environments is a field of growing interest due to its wide application in intelligent systems, such as indoor navigation, logistics management, and patient monitoring. However, traditional methods for constructing fingerprint maps, based on extensive measurements, face challenges related to the high cost of data collection and maintenance. This thesis proposes an innovative method that combines image super-resolution and data fusion techniques to optimize fingerprint construction, significantly reducing the collection effort. The study integrates measurements from different technologies, such as Wireless Fidelity (Wi-Fi) networks and Frequency Modulation (FM) radio stations, to improve the robustness and accuracy of the system in scenarios where data sources are limited or inconsistent. The methodology involves the use of previously trained Convolutional Neural Network (CNN) to process Received Signal Strength Indicator (RSSI) data, allowing the modeling and prediction of signal propagation in greater detail. In addition, data fusion techniques are applied to integrate information from multiple sources, maximizing localization reliability under challenging conditions. The experiments were conducted on 716 reference points in a controlled environment, with comparative performance evaluation between traditional methods and the proposed model. The results demonstrated a reduction in data collection effort, while maintaining high location accuracy, with an average error close to 1.31 meters using less than half of the pre-defined Wi-Fi networks. It is concluded that the proposed approach offers an efficient solution for the construction of fingerprint maps, with potential for application in several practical scenarios. This study advances the state of the art in indoor and semi-open localization, opening new paths for the use of deep learning and data combination in localization systems.