Sistema híbrido de localização indoor baseado em RFID e análise visual

Abstract

Systems need to know the physical location of objects and people in order to improve the user experience and solve logistic and security problems. This work proposes a hybrid indoor positioning system based on passive RFID and visual analysis. The system focuses on bidimensional and tridimensional space localization of stationary objects with centimeter level accuracy. Also, off-the-shelf equipment are employed as sensor devices. By using RFID technology, machine learning approaches based on Artificial Neural Networks (ANN) and Support Vector Regression (SVR) are proposed. A Computer Vision (CV) subsystem detects visual markers in the scenario to enhance RFID localization. In order to combine RFID and optical technologies, we propose a novel sensor fusion method based on Multiple Region of Interest (MROI) and k-means technique. A multi-frequency method is proposed aimed to allow and improve the localization when using off-the-shelf equipment. We have implemented our system and evaluated it using real experiments. First, a performance benchmark was made in order to evaluate RFID antennas and tags positioning. Regarding to overall system performance, the localization error was between 9 and 33 cm under a 2D scenario. On this dimension, ANN performed 30% better than RNA approach. In comparison to RFID-only approach results, the hybrid system had improved by 32%. Three-dimensional space localization had a 63 cm accuracy in best case scenario.