Your CCTV already watches every entrance. Glimpse turns those cameras into accurate footfall counters, and then into everything else the same cameras can see: passers-by, dwell, flow, queues and heatmaps.
This page explains how CCTV footfall counting works, what makes a camera suitable, how accurate it is, and how it compares with overhead sensors.
Trusted by the world's biggest retailers, brands, transport operators and property groups
A Glimpse processing device on your network receives the live RTSP stream from each suitable camera, via your NVR or directly from the camera. Nothing about the camera changes; nothing is re-cabled.
On the device, a computer vision model finds each person in the frame and reduces them to an anonymous position - a box, not a face. The video frame is processed and discarded in the same moment; it is never stored and never transmitted.
A temporary, anonymous ID follows each position across the view. A virtual footfall line counts each person once in each direction for accurate entry and exit counts.
In most retail estates, yes, and the true answer for any particular entrance is "we'll tell you after we've looked". The question has two halves: the view and the camera.
A good view has three properties. It is clear: the camera sees the whole width of the entrance you want to count, with a little margin either side, and nothing (e.g. a pillar, a display, a door that opens into shot) hides part of it. It shows the complete height of the doorway: the camera should see people from head to foot as they cross the line, because the model uses the whole body to decide that a shape is a person, where their feet are, and which way they are going. A camera that cuts people off at the waist as they enter counts worse than one that sees all of them. And it has low occlusion: people crossing the line should be visible as individuals, which usually means a camera mounted high enough to look down towards the entrance rather than across it. Most door cameras fitted for security already have all three properties.
The camera itself matters less than people expect. Glimpse works with any camera that can provide an RTSP stream to a device on your network - which is virtually every IP CCTV camera made in the last 15 years, from every major manufacturer, either directly or through your NVR. Standard HD resolution is more than enough: the model needs a person to be a reasonable size in the frame, not a sharp picture of them. The things that limit a feed are a very low resolution, poor lighting conditions that limit contrast, or an analogue camera with no network output, and even those usually have a route, either through an encoder or the NVR.
An overhead sensor counts the strip of floor beneath it. To move the counting point you have to move the device - an engineer visit and disruption to your store. A footfall line in Glimpse is drawn in software over the camera view, so it can go anywhere in that view, and it can be moved, added to or duplicated as a configuration change with no one on site.
That changes where you can count. Put the line where the visit actually begins: past a lobby where people shelter from the rain, beyond an ATM vestibule, after a reception desk in an office or a showroom, so that the count is people who came in rather than people who stood in the doorway. Draw a second line across the entrance to a department, a fitting room or a concession and the same camera gives you flow - how many of the people who came in got that far. Draw a line along the pavement in a frontage camera and you have passers-by, the denominator for capture rate. Draw a polygon instead of a line and you have a zone, for dwell, occupancy and queues.
Any number of lines and zones can be defined, each tagged by category - footfall, passer-by, flow - and aggregated up the hierarchy from an entrance to a store to a region to the estate.
