Lab notes · Queue analytics
Why airport queue counts run low — and how a top-down camera fixes it
Queue and crowd counts in busy places like airport security lines often fail for one simple reason: people block each other from the camera's view.
Under standard angled cameras, up to 67% of tracked people overlap. That makes the count read low — exactly when the line is longest, which is the moment it matters most.
We tested our pipeline with a true top-down (nadir) camera to measure how camera angle affects accuracy. Looking straight down greatly reduced heavy overlap, even in dense crowds.
There was a catch. Standard detectors, like stock YOLO models, look for upright human shapes. From directly overhead, a person is just a head-and-shoulders outline, so a standard model found only a few people in a crowd of hundreds. Pairing the top-down view with a model trained on overhead images found about 300 people per frame and reliably tracked more than 1,580 different people.
The result is the numbers an operations team can act on: head count, throughput, walking speed, estimated wait, dwell time and a density heat map.
Accuracy depends on camera physics as much as on the model. Get the camera angle right first.
Capability demo built by IPD Scientific. More lab notes →
Have footage like this?
Book a free 15-minute call, or email us a sample image or short clip. We'll tell you honestly whether computer vision can do the job.