Why Crowd / People Counting Tracker Works Here
Detects and boxes every person
Object Detection's COCO-SSD model recognizes people among its 80 trained classes, live.
Numbered labels
Turn on Labels to number each tracked person, or show detection confidence per box.
Adjustable sensitivity
Score Threshold controls how confident a detection needs to be before it counts — tune for a crowded or sparse scene.
Duplicate suppression built in
Overlapping duplicate boxes for the same person are automatically suppressed.
Three model accuracy tiers
Lite, Standard, or Heavy model accuracy, trading speed for precision depending on your device.
Works on recorded footage
Analyze an uploaded video instead of a live feed for the same person-counting tracking.
Frequently Asked Questions
Can it count people crossing a line?
Yes. Turn on Counting Line, drag a line across the frame, and it tallies crossings in each direction separately — footfall through a doorway, or people passing a point.
How does it avoid counting the same person twice?
Counting is keyed to the tracked ID, and a crossing only registers when that track actually passes through the drawn segment. One person crossing once counts once, however many frames they take, and traffic passing beyond the ends of the line is not counted.
Can I count only part of the frame?
Not with a drawn region in Object Detection mode — zone restriction (Add Zone / Only Track Inside Zones) is a Motion Tracking mode feature. The Counting Line above already covers most doorway/threshold counting use cases, since only crossings of that line get tallied regardless of where else in frame people appear. For a full "ignore everything outside this box" zone with a live occupancy count, switch to Motion mode — its own person detection is weaker than Object Detection's, but it has zones.
Can I ignore people or objects that are standing still?
Yes. Ignore Stationary Objects drops anything that has not moved across its trail history, and Min Speed hides anything drifting slower than a threshold you set — useful for excluding a queue that is not moving.
How many people can it handle at once?
There is no fixed cap. Accuracy depends on how much people overlap — a busy crowd where bodies heavily occlude each other will merge or miss detections, so an overhead or angled view counts far more reliably than a head-on one.
