experiences with AI analytics

[SJS]

I am currently working on my bachelors thesis about the potentials of AI Video Analytics for event scenarios. 
I am specifically interested in how well they work and the false alarm rates. 

It would help me alot, if anyone has some numbers or experiences to share.

Thanks and Regards

Edited December 18, 2025 by SJS

[samuelharris]

Specific false-alarm numbers are hard to generalize, since results depend heavily on camera placement, lighting, crowd density and how the vendor defines an event. For a thesis, published benchmark datasets are usually more defensible than anecdotal figures, because deployment reports rarely state their test conditions. Ask vendors for precision/recall at a stated operating threshold rather than a single "accuracy" number, and check whether the reported rates were measured on-site or in demo footage. Also worth noting that event venues vary a lot in density and occlusion, so any single figure deserves caveats.

[juliawhite]

Building on the point about test conditions, the on-site versus demo footage distinction matters a lot for your thesis. Demo clips tend to be cleaner and better framed than real venue feeds with variable lighting and heavy occlusion. Asking vendors for precision and recall at a stated threshold gives you something comparable, but I'd pair that with a small set of published benchmark results and explain the gap between controlled datasets and event deployment. Being explicit about those caveats will make your conclusions more credible than a single averaged figure.