Recently, the indoor positioning schemes have been actively studied. The indoor positioning scheme can be roughly classified into four schemes using triangulation, fingerprint map, proximity or visualization, respectively. This paper introduces the preliminary cut-off indoor positioning scheme using the fingerprint map.
The preliminary cut-off scheme improves the accuracy of the K-Nearest-Neighbor (KNN) algorithm, which is a typical scheme of using the fingerprint map. This scheme consists of two phases: off-line and on-line phases. The off-line phase constitutes a fingerprint map necessary for real-time positioning. APs that periodically generate a signal are arranged and reference points predefined in indoor environment are selected. Then, the RSSIs received from nearby beacons at each reference point are stored in the fingerprint map. The on-line phase actually estimates indoor position. The user’s device receives the signal of the nearby APs. User’s position is estimated by comparing the RSSI received in real-time and RSSI stored in the fingerprint map.
The KNN algorithm uses the Euclidean distance to compare the RSSI received in real-time and the RSSI stored in the fingerprint map. The K reference points with the shortest distance are selected and the position of user is estimated as the center of these reference points. However, since there are many obstacles in the indoor environment, the strength of signal is not constant even at the same position. To mitigate the instability and variability of the radio signal, the preliminary cut-off scheme utilizes the relative rank of the peak of signal strength, not the signal strength. Then, the user’s position is estimated as the center of K reference points with the greatest similarity after calculating the similarity between the real-time ranking and the ranking of the fingerprint map.
This paper describes a continually improved study to improve the accuracy of the preliminary cut-off scheme. As a result, not only similarity to the relative rank of the peak of the signal strength but also the similarity to the peak is calculated and a weight based on this similarity is assigned. The user’s position is estimated to be the calculated position by weighting each reference position, not the center of the K reference positions.