A Novel Clustering Algorithm for Wi-Fi Indoor Positioning

• Published in: IEEE access Vol. 7; p. 1
• Main Authors: Ren, Jin, Wang, Yunan, Niu, Changliu, Song, Wei, Huang, Songyang
• Format: Journal Article
• Language: English
• Published: Piscataway IEEE 01.01.2019; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Accuracy; Algorithms; Classification algorithms; Cluster analysis; Clustering; Clustering algorithms; Euclidean distance; Fingerprint recognition; Fingerprints; Improved public c-means (IPC) clustering algorithm; Indoor environments; Indoor positioning; K-nearest neighbors algorithm; Received signal strength indicator; Signal strength; the k-nearest neighbors (KNN) algorithm; The knearest neighbors (KNN) algorithm; Vector quantization; Wi-Fi; Wireless fidelity
• DOI: 10.1109/ACCESS.2019.2937464
• ISSN: 2169-3536

In recent years, the Wi-Fi-based indoor positioning technology has become a research hotspot. This technology mainly locates the indoor Wi-Fi based on the received signal strength indicator (RSSI) signals. The most popular Wi-Fi positioning algorithm is the k-nearest neighbors (KNN) algorithm. Due to the excessive amount of RSSI data, clustering algorithms are generally adopted to classify the data before KNN positioning. However, the traditional clustering algorithms cannot maintain data integrity after the classification. To solve the problem, this paper puts forward an improved public c-means (IPC) clustering algorithm with high accuracy in indoor environment, and uses the algorithm to optimize the fingerprint database. After being trained in the database, all fingerprint points were divided into several classes. Then, the range of each class was determined by comparing the cluster centers. To optimize the clustering effect, the points in the border area between two classes were allocated to these classes simultaneously, pushing up the positioning accuracy in this area. The experimental results show that the IPC clustering algorithm achieved better accuracy with lighter computing load than FCM clustering and k-means clustering, and could be coupled with KNN or FS-KNN to achieve good positioning effect.