Person Detection on H7

Hi, I have just purchased the H7 board plus a couple of lens options and have got it running successfully but I have a problem with the person detection function operation.
It seems to basically work OK and will detect a person in the FOV but I am getting a lot of false triggers off objects in the room especially under different lighting conditions when no person is present.
I am new to this and was wondering if there is anything I am able to do to eliminate these false triggers off other objects that aren’t human?
I have an application that requires detecting a human within about 2m of a moving vehicle. I have tried LiDAR and IR but there are problems using these technologies and thought I would try machine vision to solve the problem of distinguishing between a solid object and a human. Is this possible with OpenMV?

Yes, what’s the confidence score you are looking at? It tells you the percentage (0-100) it thinks there’s a person there. Just raise the threshold. It might be set low right now.

Yes, the person detection should work for that.

Hi Kwagyeman,
Yes I have experimented with the thresholds by raising them up to 85% -90% as well as combining the other thresholds in different combinations but I am still getting a lot of false readings as well as missed detections.
I am finding that the environmental lighting has to be very consistent as one set of threshold settings will work in the morning and won’t work in the afternoon when the daylight coming into the room is greater. I have also noticed this problem when switching on lights in the room at night.

You may wish to use a temporal filter to remove false positives. Note, the person detector network is Google’s visual wake words network for person detection.