The Measurement Minute by Gary Angel

 

Ghosting, stitching and matrixing are the three big gotchas when you try to use people-counting cameras for more than door-counting. It’s true that people-counting video cameras are better than 99% accurate at door-counting applications – but that doesn’t mean they are 99% accurate at people tracking. Why not? It turns out the that the ML in people counting cameras is highly optimized for door-counting and often breaks in other situations. In this Measurement Minute, I describe the three biggest problems and their partial solutions.