A fingerprint-based coarse-to-fine algorithm for indoor positioning system using Bluetooth Low Energy

In this research, we propose a fingerprint-based fine-tuning algorithm to increase the accuracy of indoor positioning system using Bluetooth Low Energy fingerprint. In the fine-tuning step, Delta rule is used to update the coordinates of reference points based on the wireless environment represented in the training dataset. The combination of the coarse estimation and the proposed fine-tuning makes the coarse-to-fine algorithm. The algorithms used for coarse estimation are weighted sum and k-nearest neighbour, with three weight calculation algorithms that are compared in this experiment: Minkowski distance, k-means Gaussian mixture model, and autoencoder. Two subject rooms are used in order to measure the performance of the algorithms. The experiment showed that the fine-tuning algorithm improves the accuracy of the positioning throughout all combination of methods in both rooms, which shows its versatility. It reduces the mean positioning error by up to 11.3% and depends on what algorithm used in the weight calculation. Another benefit from the fine-tuning model is that it does not increase the complexity of the algorithm in the online phase. Overall, the best result is achieved by the combination of weighted sum and k-nearest neighbour with Minkowski distance weight calculation, together with the proposed fine-tuning. Its mean positioning error is 0.8740 m for Room 1 and 1.5385 for Room 2. The average computing time for a single online query is 0.4 ms for Room 1 and 0.7 ms for Room 2.