What smart detection is good and bad at
On-camera analytics have become genuinely useful for one thing: cutting false alerts. Person, vehicle and animal classification means your phone stops buzzing every time a shadow crosses the lawn or snow slides off the roof, and that single change is what makes people keep notifications switched on.
Beyond classification, be sceptical. Line-crossing and zone intrusion rules work well when drawn carefully and badly when drawn hopefully. Facial recognition and familiar-face features vary widely in accuracy, raise their own privacy questions for visitors and staff, and should be a deliberate choice rather than a default. Package detection is handy. Anything marketed as predicting intent is not something to buy a system for.
We tune the detection zones and rules on site, in the light and weather you actually have, then revisit them once you have lived with the alerts for a week.