AI motion detection with BlueIris: an alternate approach

Did you read the section on high cpu usage?
 
Did you read the section on high cpu usage?
Yes, but I read it again. I've changed the BI setting to only post a jpg in the AIINPUT directory when it sees motion so now I have to wait for an animal or somthing to trigger a jpg post (the system is remote and I can't just walk in front of one of the cameras).

Thanks again for your help.
 
I'm having a strange problem. After I had to reinstall docker, my setup runs for about 15-20 hours then stops processing images. The containers are running, there are no errors in the logs, but they just stop processing anything. On restart it works again. Having a real hard time diagnosing this one.

I just restarted my server, but realized I should have checked CPU util and more importantly memory usage. I'll do that next time and see if I can find a correlation.
 
I'm having a strange problem. After I had to reinstall docker, my setup runs for about 15-20 hours then stops processing images. The containers are running, there are no errors in the logs, but they just stop processing anything. On restart it works again. Having a real hard time diagnosing this one.
This is Docker on Windows I'm assuming? I was having this problem as well only mine would stop processing after only 6 or so hours. The "fix" that is working so far is to give it much less work to do by sending images only when a low resolution camera detects motion (these SD cameras are set to a very high sensitivity) and thus the AI engine is only having to analyze about a half dozen images per hour.
 
I see. Yes, running on Windows docker. Funny thing is the PC is way overpowered, just a i9 gaming PC that I wasn't using, so even with BI running, it doesn't ever get too taxed.

I figured that I'd tell BI to also not save past a day of images in the directory, I had it set at 3 days, but that's a lot of image files in one directory...

I do currently have all of cameras, which are set to SD, sending images on motion, I have the cameras cloned and only record alerts if the deepstack config verifies motion. But maybe I can try to up the detection thresholds to send fewer images.

Thanks for the idea.
 
I highly recommend setting BI to clear the folder after 1 hr or 1GB of data. The more images you have in the folder the slower things will get due to how new file detection works.
 
I see. Yes, running on Windows docker. Funny thing is the PC is way overpowered, just a i9 gaming PC that I wasn't using, so even with BI running, it doesn't ever get too taxed.

I figured that I'd tell BI to also not save past a day of images in the directory, I had it set at 3 days, but that's a lot of image files in one directory...

I do currently have all of cameras, which are set to SD, sending images on motion, I have the cameras cloned and only record alerts if the deepstack config verifies motion. But maybe I can try to up the detection thresholds to send fewer images.

Thanks for the idea.
Just to be clear, I have set the motion detection on the cloned or SD cameras to be very sensitive which will produce false detections but the AI engine will sort that out and only send a trigger to the HD version of the camera when one of the trigger definitions (person, dog, truck, car, etc...) is fired. This may send a lot of false motion images during fog, rain, snow, etc,... but it will be FAR LESS images then when I was having each clone SD camera send an image every 5 seconds regardless of motion or not which was overwhelming the AI engine in the Docker container and making it unstable / unreliable. And yes, I have BI keeping only an hour's worth of files in the AIINPUT directory.
 
Oh yeah, sorry I wasn't clear that's exactly how I have mine set. Only alerts send images from the SD cams to Deepstack. However I have 15 cameras or something so quite a few images make there way over to deepstack. I should do an inventory of how many images are delivered at night which is when it seems to die most often, maybe too many images are going out too quickly.
 
Another thing you can do is look at splitting the camera images out into separate folders. Have a folder for, say, every 5 cameras, then update the watchPattern on the cameras to point to the correct sub-folder.
 
Yes, but in real world setups you don't get that many notifications, so chances are the clip you want is the first clip in the list.
 
Hallo is it possible to store an annotateImage in a extra directory?

If i get a positive detecten of an interesting object, i would ike to store the Image with annotate of the object in a extra directory (Storeing them for same days)
It would be helpful to check why sometimes an object is detecting which i can't find in the video.