Projection-induced Access Point Deployment for Fingerprint-based Indoor Positioning

Location information and positioning technology are important to many of the emerging Internet of Things (IoT) applications, and WLAN-based positioning is one of the promising solutions due to the prevalence of Access Points (APs). Nevertheless, different deployments of APs may have significant impact on positioning performance as it may generate different distribution of signal features in surrounding environments, which form the basis of fingerprint-based localization. Current efforts on AP deployment mainly focused on enlarging the probabilities of small errors while ignoring the probabilities of big errors. However, big error could significantly affect the user's experience so that it should be paid more attention. Therefore, in this work, we propose a projection-induced AP deployment approach, whose principle is decreasing the probabilities of big errors. Specifically, firstly, when constructing objective function, unlike the conventional approaches which considered all Received Signal Strength (RSS) vectors collected in every Reference Point (RP), we do outlier detection using K-Nearest Neighbors (KNN) graph previously. Secondly, we solve the defined objective function from the projection perspective rather than search algorithms, which would bring computing consumption with iterations. Finally, we build the system prototype and implement in our environment and the experimental results demonstrate the effectiveness and the efficiency of the proposed AP deployment solution.