What is sensor calibration?

The aim of the sensor calibration is to deliver accurate measurement of occupancy & movement.

One challenge of counting the number of phones in a space is that spaces and phones vary. Some locations are constructed of materials that block Wi-Fi signals, while others allow Wi-Fi signals to pass freely. Some spaces have internal obstacles (e.g. metal filing cabinets), while others do not. And few spaces are circular, a case that would allow signal strength to be a good proxy for the perimeter of the area. Some phones probe more aggressively and with higher power than other phones… though the variation among phones is surprisingly small.

As a result, BlueFox sensors operate best if they are calibrated, a process that compares “ground truth”, the true number of persons in the detection zone, with the Wi-Fi signals collected by BlueFox sensors. Generally, we like to collect ground truth on several different occasions, both when the space is crowded, and when it is less crowded. We offer a purpose-built application, BlueZoo Foot Traffic Analytics mobile app, that permits a group of people to simultaneously collect ground truth, for example, at different entrances to a space. Our systems collate this ground truth and compare it to the values detected by our sensors to calibrate sensors for optimal sensitivity.

One technique that BlueZoo employs is to collect a series of entrance/exit data. We don’t require an exact occupancy count at the beginning of the entrance/exit series. Staff with our mobile app are placed at each entrance/exit and count the number of persons entering and leaving the space. Our mobile app transmits this information to the BlueZoo cloud where we extract in/out traffic data for the time period when all entrances are covered.

First, we select one of our half-dozen computational methods that we have honed since 2016 and which are protected by trade secrets. We apply this computational method to the ground truth data to minimize the mean square error across all sample points.

Second, in the statistical counterpart of curve-fitting, we match the exact entrance/exit data with sensor data, scaling the data to achieve a good fit, and deducing the ground truth at the outset of the measured entrance/exit sequence.