Free weekly demonstrations. Register for a demo

Addressing Privacy Concerns with People Counters

April 15, 2021

A concern we hear often from those considering a people counting system is,
“Well if the sensors are video-based, isn’t that a privacy issue?”

sensource patron counting solutions

First, it’s important to clarify that people counting sensors are not security cameras. No video feed is recorded to a DVR. Video can be extracted in a small segment, which will be explained shortly. And the video quality is far from high-definition.

The device works by using two stereo-video lenses to build a height map of the detection area. It’s mounted to the ceiling and looks straight down at the floor. The brains of the operation is not the video feed itself, rather a complex string of algorithms that look for the shape and movement to detect people.

As seen in the above YouTube video, when the sensor detects a human the software places a bubble on the head and tracks the person as they move through the detection area. This allows it to capture “entrances” and “exits” as well as more advanced features such as queue metrics and dwell length.

The video feed can be accessed by SenSource’s technical support or your system admin to schedule a recording. Upon installation, the video feed is reviewed to confirm accuracy and placement. Should accuracy fall below our 97% guarantee, an hour or less video segment will be scheduled to review and recalibrate the sensor. As no video feed is stored onboard, you cannot go back in time and record or view past video footage.

The sensor is equipped with these privacy filters:

Privacy Level 0
Privacy Level 1
Privacy Level 2
Accordian image
Accordian image
Accordian image

*Setting your sensor to Privacy Level 2 eliminates the ability to certify accuracy. Accuracy validation is performed by comparing the recorded video feed with the processor’s counts. If no video feed is available, our technical support team is unable to validate the counts. For example, a sensor may be under-counting because a display table was placed in the detection area on a designated-virtual “count line”. Without the video feed, we couldn’t identify the table placement as the problem or precisely redraw the “count lines”.