Indoor localization of resource-constrained IoT devices using wi-fi fingerprinting and convolutional neural network

Location information is vital in this era of the Internet of Things (IoT). Outdoor localization has improved significantly due to advancements in satellite systems. However, the inadequacy of satellite signals in complex indoor environments has made indoor localization still a challenge. In recent years, Wi-Fi fingerprinting with deep learning has been utilized for indoor localization in multistorey buildings due to cost-effectiveness and acceptable accuracy. Its implementation on resource-constrained IoT devices with limited computing capabilities requires investigation of suitable preprocessing techniques. This paper reviews the features of publicly available datasets on WiFi fingerprinting and utilizes three datasets to compare the effectiveness of various preprocessing techniques along with Convolutional Neural Networks (CNN) that can be implemented on resource-constrained IoT devices for floor-level localization in an edge computing paradigm. Our results show up to 94.33% floor level localization accuracy with a 15.47% increment on the UJIIndoorLoc dataset when the non-detected access point's received signal strength indicator (RSSI) artificial value was changed to 1 dBm below the lowest RSSI value in the whole dataset followed by min-max normalization. Successful implementation of indoor localization in resource-constrained IoT devices has the potential to advance various sectors such as smart cities, sustainable buildings, healthcare, industrial automation, robotics and more.