People recognition problem

[denchik]

Hello!

I have a many false alarms from people detector.

My settings:
Look for: people with ignoring objects smaller than 40px
That are: inside my region
If seen: send email
My test cameras: 4pcs Acumen AiP-M53 with 1920x1080, h264, 15 fps, 5Mbps VBR video stream.
High performance Intel Core i7 PC.
Camera works in day light, no reflections.

Very often my dog detected as a people. Sometimes I recieve email alarms with screenshots, that doesn't contains any bounding boxes and no any objects. I already tried different setups of regions, objects sizes and others, but still have a many false alerts.

Maby you have some hidden settings or tunes for optimizing people recognition? Or maby you can take some advice for me how to decrease false alarms?

Thank you!

[del0]

I have the same problem just like others:
topic13072
topic12953

Almost every car in our driveway gets detected as person (although daylight, good bandwidth, "only" hd resolution).

Also Plants, spider web, birds etc.

[fester]

Waving flags too, very patriotic.

[kgonzalez]

Unfortunately, no algorithm is perfect. Sighthound Video does leagues better in both general object and person detection, but there will always be flaws due to the endless variety of scene conditions and objects. We are continually working to improve our algorithms, and hopefully, we'll be able to enhance the accuracy of your scenes in future updates. Sighthound 6.0 is under testing and development. We hope to have this update released by the end of the year. I'm afraid there are no changes we can make at this moment to improve the algorithms for your specific scenes. The following link gives tips on how to enhance detection accuracy.

https://www.sighthound.com/support/tips-getting-best-recognition

[faithhines]

Did you solve this problem successfully?