Qualitative Comparison of Existing Indoor Positioning Algorithms for BLE

  1. An Indoor Positioning Algorithm Using Bluetooth Low Energy RSSI Chai 2016
    Triangulation with RSSI pre-processing. 0.2~0.4m accuracy.
  2. Indoor Positioning System Using Euclidean Distance Correction Algorithm with Bluetooth Low Energy Beacon Wang 2016
    Fingerprinting with Euclidian distance correction. 1.58m accuracy.
  3. Location Fingerprinting With Bluetooth Low Energy Beacons Faragher 2015
    Fingerprinting. <2.6m accuracy
  4. Smartphone-Based Indoor Localization with Bluetooth Low Energy Beacons Zhuang 2016
    Fingerprinting with polynomial regression, Kalman filtering, outlier detection, and multi-channel sampling. <2.56m accuracy
  5. An Analysis of the Accuracy of Bluetooth Low Energy for Indoor Positioning Applications Jianyong 2014
    Fingerprinting with multi path mitigation. <2.6m accuracy  

  6. Improving Indoor Localization Using Bluetooth Low Energy Beacons Kriz 2016

  7. Experiences with using iBeacons for Indoor Positioning Deepesh 2016 
    This paper is an analysis of the realities of implementing Bluetooth Low Energy iBeacons (Apple) in a confined space for indoor positioning.  It examines two techniques, trilateration and fingerprinting, based on the RSSI of the Bluetooth Beacons.  The researchers built a custom app on an Apple iPad 2 to perform the experiments and check their positioning methods against the device's actual position.
    The laboratory setup was a rectangular room divided into 18 sections, 6 across and 3 down.  iBeacons were placed in the corners of the room and all fixed to the same height.  There were three pillars in the room that created obstacles for the BLE signals.
    Trilateration was first used, but abandoned quickly as the researchers found that there was large variations in the RSSI of the iBeacons.  The variations made triangulating the position very challenging and no data was obtained.
    Fingerprinting was used next and data was gathered.  Offline training data was collected by the device and then fed into two different machine learning algorithms.  The first was a k Nearest Neighbor algorithm.  The researchers found that with this algorithm, the location of the device was placed within the correct section of the room 62.7% of the time.  This indicates that there was an average of 2-3 meters shift in radius from the actual position of the device.
    The second algorithm tested is the Random Forest Algorithm, using the average solution from hundreds of decision trees to position the device.  This solution was able to achieve a correct section position 79% of the time.
    Once the tests were concluded, the researchers found a few key insights to indoor positioning.  There are vast variations in the RSSI readings of BLE signals, making the error of the positioning algorithms tested relatively high.  The device moving around, other people near the device, fixed obstacles near the device, the presence of other devices near the device, and orientation of the device and iBeacons make large, unpredictable changes to the RSSI.  These variances create a huge challenge for implementing BLE indoor positioning in crowded and dynamic environments like shopping malls and airports.

  8. An Analysis of the Accuracy of Bluetooth Low Energy for Indoor Positioning Application