Home About Releases Papers Contact-Us
| Motivation | Result |
|---|---|
 |
 |
Places like retail stores, museums, libraries are gradually moving towards a more automated experinece for customers. Amazon Go serves as an example where customers can do automated checkout of their items. Towards this end, tracking the activity of the customers in such spaces becomes an important problem that needs to be solved with high accuracy.
Finding by counting is a system which can currently localize moving customers with decimeter level accuracy.
Finding by counting supports the localization of moving customers with the following features:
- Finding by Counting can find physical co-ordinates of moving customers at regular intervals with an average accuracy of 0.4 meters.
- Finding by Counting uses BLE beacons as anchor nodes for localization.
- Most techniques use RSS or strength values of signals from fixed devices like WiFi APs and then map that to distance either through fingerprinting or by deriving a mathamatical model of signal strength decay with distance. Finding by Couting ignores RSS value altogether due to its high variance and unreliability in indoor settings. It simply counts the number of packets received from the beacons.
In our work, we estimate distance by counting the number of packets received from stationary beacons.
- We train a Generalized Linear Model(GLM) on fraction of packets received from beacons. GLM contains distance, beaconing power and advertising frequency of beacons as parameters.
- We use trained GLM in a range free localization setting i.e. along with a Kalman filter to infer location of a moving person over subsequent time windows.
- We use Markov Chain Monte Carlo (MCMC) techniques to learn GLM parameters during training phase and location parameters during localization phase.
| Trained Model | Accuracy Comparison |
|---|---|
 |
 |
Code and Data are available here
Finding by Counting: A Probabilistic Packet Count Model for Indoor Localization in BLE Environments
Subham De, Shreyans Chowdhary, Aniket Shirke, Yat Long Lo, Robin Kravets and Hari Sundaram, WiNTECH, Mobicom'17, Snowbird, UT, USA. October 2017
Finding by Counting is being developed by a team of undergraduate and graduate students headed by Prof. Hari Sundaram along with Prof. Robin Kravets. The list of contributors includes: Subham De, Shreyans Chowdhary, Aniket Shirke, Yat Long Lo.
Please reach out to the lead PhD student, Subham De ([email protected]) if you'd like to either contribute, or be a tester of Finding by Counting!