false positives on shadows

[jeremyclark]

On sunny days, I tend to get a barrage of false positives triggered by shadows with the settings I am using. Below are the settings I have and an example of what triggers it. I'm afraid to make the the exclusion of regions larger than 60 pixels (and it only goes to 70) as that's about the size of a person who might be walking up my driveway. Since I have a number of large oak trees surrounding my house, they tend to cast shadows that show up in all of my various cameras. On a sunny, windy day, this can trigger 20 emails within a couple minutes at times, and it can go on for hours. When this happens I pretty much have to turn off the rules, and then inevitably I forget to turn them back on and wonder a few days later why I haven't been getting alerts when people show up at my door, etc.

Any suggestions? Or Ryan, is there anything that could be done in the detection software to improve shadows from triggering them (perhaps detecting when it's within a certain range of colors that would essentially just mean a slightly darker color than it was a few seconds ago)? Or a way to adjust the sensitivity with the extreme being very different colors in a certain region?

Screen Shot 2014-06-09 at 2.21.00 PM.png

(198.02 KiB) Downloaded 393 times

Screen Shot 2014-06-09 at 2.20.35 PM.png

(203.07 KiB) Downloaded 393 times

IMG_1872.JPG

(42.34 KiB) Downloaded 393 times

[Devin]

Hi jeremyclark,

Have you tried adjusting your cameras brightness and contrast settings? You may be able to eliminate the false detections by increasing the brightness (it seems that the shaded areas in your screenshots are very dark).

Using the "Crossing a boundary" option may work best for detecting people walking up or down your driveway, as it will ignore anything that did not cross your boundary line. I've attached an image of how you could set the boundary to catch any person walking down or up your driveway.

[Devin]

Hi Jeremyclark,

If you only want to only capture people on your driveway you can set the "Look for:" option to "People". This would eliminate the false "object" detections (the green boxes shown in your screenshot).

Disregard this post if you intend to have this rule triggered by vehicles as well as people.

[jeremyclark]

Thanks Devin - yes, I've tried changing that to "People" but then it doesn't pick up vehicles which just as important (if not more so) for me. I wish there was a way to "look for:" vehicles and people!

I've tried adjusting the brightness and contrast but haven't had a bright sunny day yet to test things. Hopefully it'll help.

[jeremyclark]

Plus, I enjoy being able to capture moments like this one from this morning! (although I think Sighthound did think this was a "person"!).

Garage.jpg

(25.53 KiB) Downloaded 369 times

[ryan]

Hi Jeremy,

I'm with you on enjoying capturing the wildlife clips. Though I'm afraid all I have around my place is skunks...

We're certainly working on better shadow elimination in our next generation algorithms, hopefully that will solve all of your issues. In the meantime to expand on some of the suggestions here already, are the only shadows that trigger in your scene the ones that are in the lower left of the frame? If so, note that you can arbitrarily shape your region (not just four points) so you could create a "walking on" or other region around this area, or you can make a small "outside of" region encapsulating the lower left, that would still get any object that *ever* appeared outside of that region. Examples attached (though I didn't photoshop in your scene, just consider the shapes).

[jeremyclark]

Thanks for your thoughts Ryan. Good to hear you're working on improvements for shadows!

Per some suggestions you mentioned in the past on these forums, I did discover the non-rectangular shapes and I'm using them on some of my other cameras already. Unfortunately in my situation the shadows in the driveway are all over the place, as shown in these examples:

IMG_1921.JPG

(42.31 KiB) Downloaded 362 times

IMG_1919.JPG

(39.09 KiB) Downloaded 362 times

IMG_1918.JPG

(51.08 KiB) Downloaded 362 times