Motion sensors on Sub stream profiles

singularity42

singularity42

Young grasshopper
Jan 21, 2020
46
15
MN
I moved a lot of my cameras over to sub profiles for a lot of things within BI, to decrease CPU use. I realize by doing this that motion sense also uses the sub profile and the resolution is thus lower. However, even setting my cameras motion sensors to min object size, it still seems they dont do very well detecting objects. Ive tried multiple different settings within motion sense in BI, but still cant find the right settings.
Any ideas on how to resolve this or improve it?
 
Post some screen shots of the motion detection field of view and showing the min size and min contrast settings.

Also make sure that you redrew the motion zones as they change with the substream.

Also make sure that you uncheck the object moves so many pixels and resets after 67% - if those are checked that could be causing the problem.
 
  • Like
Reactions: sebastiantombs
I did change the motion zones. I do however have the other 2 still set. Ill try disabling them. Here is an example of one of the cameras that has issues with object motion.
1635300716762.png
 
How about trying the Algorithm options?
Screenshot 2021-10-26 211803.png
 
I usually run make time down around .5-.7 seconds.
 
  • Like
Reactions: Flintstone61
Yeah, at night try simple and drop the run time.

Not sure what machine you have, but having the camera time overlay and the BI time overlay does consume some CPU.

Not enough contrast in the image that with a substream could also make it a little more difficult.
 
  • Like
Reactions: Flintstone61 and sebastiantombs
Thanks for all the feedback and replies. My question to those of you that have such low make times, how do you possibly deal with the 10's to 100's of false positive motion sensing due to insects?
My thought on the higher make times is attempting to catch a person or animal that is walking.

Not sure what machine you have, but having the camera time overlay and the BI time overlay does consume some CPU.
Realistically I could probably go back to not using sub profiles. I just bought a new DL380 G9, 24 core. With my current 6 (sometimes 7) cameras, I use about 10-25% of the CPU depending on load.
 
Many will run external IR to eliminate that.

You can get creative with zone crossings to eliminate a lot of it.

I take bug spray and spray around my cameras running IR and that knocks it out for awhile. Consider it preventative maintenance and I spray around the whole house now to help keep critters from coming inside, so I just shoot up at the camera when doing so. A couple times a year usually takes care of it for me.

You could force it into color and deal with the degradation potential due to not enough light.

You could run DeepStack to not alert you on anything other than what you want to be alerted to.
 
Last edited:
Ok, thanks. I'll just have to deal with it I guess. Interestingly, when running in main profile and not sub, I did not have any of these issues. Motion detection captured all the cats, foxes, coyotes and deer, that my wife loves to look at :) - and seldom captured insects. I cant run without IR and probably just have to deal with insects since I live in the woods. Maybe Ill look into external IR. Thanks.
 
I just live with the extra motion events. If you choose 24/7 continous and triggered, you get everything, if something actually happens, loss of property vandals etc, it'll be there, you'll just have to scrub through the video.
I can watch a whole night of parking lot in 5 minutes looking thru alerts. I break my continous video into 15 minute segments. it's easier to search that way.
 
  • Like
Reactions: sebastiantombs
wittaj said:
You could also use DeepStack to eliminate the bugs - forgot to mention that above.
Click to expand...

Right thats my next project. Although I want to be able to detect animals, as well as humans/cars.

Flintstone61 said:
I just live with the extra motion events. If you choose 24/7 continous and triggered, you get everything, if something actually happens, loss of property vandals etc, it'll be there, you'll just have to scrub through the video.
I can watch a whole night of parking lot in 5 minutes looking thru alerts. I break my continous video into 15 minute segments. it's easier to search that way.
Click to expand...

Ah! Good idea actually. I'll move mine from the 2 hours they are at, to 30 mins and see how that works for us.
 
I use DeepStack to eliminate false triggers. DeepStack will identify a raccoon as a "cat" or "dog, deer as a "dog" or "person", well you get the idea.
 
  • Like
Reactions: Flintstone61
Well you can have it detect just about anything, including a toilet and toothbrush LOL. Or train a custom model.

But the advantage to DS is you can let the cameras trigger for whatever you want, but DS can then filter out to only alert you for objects that meet your criteria.
 
  • Like
Reactions: sebastiantombs
Awesome, thanks guys. Ive got a lot of experience with tensorflow and building custom models. Been meaning to go this route, you've all given me the kick in the arse to start.
 
  • Love
Reactions: Flintstone61
You must log in or register to reply here.
Top Bottom