Novel indoor localisation using an unsupervised Wi-Fi signal clustering method

Indoor localisation continues to be an important research challenge. One interesting approach is to realise the localisation without specific additional hardware, such as special tags or ultrasonic systems, but rather with typically already available infrastructure such as Wi-Fi and cellular networks. Localisation techniques based on supervised learning, as observed in many previous investigations, require the availability of location labels that should be provided by experts or users. Unsupervised learning techniques are seen as an alternative where the localisation system automatically detects useful information to recognise locations without requiring explicit labelling from users. In this paper, a novel indoor localisation method using a density-based clustering method is presented. It utilises measured signal strength from surrounding Wi-Fi access points (APs) to automatically create fingerprints according to the accumulative frequency density of the signal strengths. The evaluations have shown that the algorithm is capable of room-level localisation with high precision in a normal office environment.