My holiday lights seem to be human! Detection help.
I've reduced sensitivity quite a bit. But My holiday lights are still triggering "human detection". I mean, I'm only 158 pounds at 5' 11", and look like a closed umbrella when I wear black... But I think hanging lights could use an AI tweak for detection. I haven't met a person that looks like them, yet. LOL!
Any other thought about this? I'd hate to mask the whole spot. Not going to be an issue soon, but it has been for 3 years now.
Comments
-
Sometimes when the AI is faced with some particular objects that don't feature in its training data, the results can be unexpected. However, what I think could be happening here has a more mundane explanation: when you have hanging objects like this, when it's even a bit windy they will sway constantly and cause a huge number of motion triggers. SecuritySpy's AI detectors are very accurate (up to 99%), but even when you have this high accuracy, you will get a significant number of incorrect results when you have so many triggers. For example, if there are 1000 triggers per hour from the swaying lights (which could easily happen in this case with a bit of wind), you would expect around 10 false-positive detections, which will feel like a lot, even though the AI detector is operating at its expected 99% accuracy.
The things to do here are:
- Get rid of the hanging lights. This will solve the problem. Instead, use string lights that don't hang down.
- Under Settings > Cameras > Triggers > Advanced Options, make sure the trigger duration is set to 1.5s or longer.
- In that same settings panel, you could increase the AI detection thresholds (e.g. to 80). This will reduce the rate of false-positive detections, but it will have the effect of increasing the rate of false-negative detections (missed events).
-
Hi Ben
Seems like "wind" is an Achilles Heal for detecting correctly anything in SecuritySpy. I have been fighting it since I pointed the cameras outside with surrounding trees. If it isn''t the branches its shadows casted from the branches. I fight the snow and wind every spring and winter. The only AI that completely works seems to be me going through the captures and discarding the giga byte files that have been created. So the question is what is a really good work around for "even a bit windy". Almost seems that in order to save file space even a "May ham filter" would help (Motion over a large portion of the frame would be discounted) . I don't know - I do feel the pain
Dale
-
Thanks! I'll give all that a try. I can still pick up cats and racoon, so sensitivity can probably be a bit less.
-
I do have issues with tree shadows when the sun is low, as during winter. All I can make out for detection is the shadows are moving across the ground as the sun moves. Shows up as vehicle detection, with the cars parked in a nearby driveway.
-
@Ldalep - extraneous motion like branches and shadows is going to be a challenge for any motion-detection algorithm, whose job is to, well, detect motion. This is precisely what the AI is designed to resolve: it should be weeding out the vast majority of these false-positive detections, so that very few will cause a recording or a notification. If this isn't work well for you, then perhaps we can help you find some settings that will make it work better - please email us and include an image from the affected camera(s) as well as a debug file (SecuritySpy menu > Debug > Create Debug File On Desktop) and we will advise you further.
