Indoor Positioning System using Regression-based Fingerprint Method

Indoor Positioning System has opportunity to be used in different business platform. Based on past research, optimized localization method for Bluetooth Low Energy (BLE) to predict position of person or object with high accuracy has not been found yet. Most recent research that have solve Received Signal Strength (RSS) inconsistent value is using fingerprint method. This paper proposed a deep regression machine learning using convolutional neural network (CNN) with regression-based fingerprint model to estimate real position. The model used 5 nearest fingerprints as reference RSS values with their location (x or y) label as inputs to produce output of single value position (x or y), then repeat the process to produce second value of position to create complete coordinate of estimated position. To evaluate the proposed model, a comparison between training data with validation data using Root Mean Squared Error (RMSE) is used. The comparisons are with Multilayer Perceptron model and with the weighted sum method as benchmark. The experiment Gave results of mean distance and 90th percentile distance between proposed model with the benchmark. CNN model achieved accuracies of lower than 330cm at 90th percentile with mean distance lower than 185cm. Weighted sum model achieved accuracies lower than 360cm at 90th percentile with mean distance higher than 185cm, and MLP is in between them. The result demonstrates that the proposed method outperformed the benchmark methods.

@article{Ghozali2019,
  title     = {Indoor Positioning System using Regression-based Fingerprint Method},
  journal   = {International Journal of Advanced Computer Science and Applications},
  volume    = {10},
  number    = {8},
  year      = {2019},
  publisher = {The Science and Information Organization},
  author    = {Reginald Putra Ghozali and Gede Putra Kusuma},
  doi       = {10.14569/IJACSA.2019.0100829},
  url       = {https://doi.org/10.14569/IJACSA.2019.0100829}
}