BI stops detecting motion randomly

goldbond

n3wb
Joined
Jul 16, 2020
Messages
23
Reaction score
14
Location
Atlanta
BI 5.4.3.10 (as of today), i7-9700, 32 gb ram, WD 4TB blue, RTX3070 with only 3 Amcrest (1x Ultra 4K and 2x 5MP) cameras . Note: I am using AI. However, it has nothign to do with the issue. It doesn't get that far..

BI randomly stop picking up motion. I mean it is broad daylight EST at the moment. And, it decided it didn't want to notice a huge red pickup truck coming up my laneway. There is not a single AIInput image for AI to pickup. This is happening multiple times a day to the extent that the software is quite honestly useless if I cant resolve. All this morning, it picked up people, cars..this is stupid..

Appreciate any insight as I am ready to throw in the towel..

NOTE: I tried deselecting object travel and size exceeds and made ZERO difference.
 

Attachments

fenderman

Staff member
Joined
Mar 9, 2014
Messages
36,897
Reaction score
21,250
You also have reset detector set way too low leave it at the default
 

goldbond

n3wb
Joined
Jul 16, 2020
Messages
23
Reaction score
14
Location
Atlanta
Change edge vector to simple
I'll try and thanks for the quick response. Can I ask why you suggest this? I did read up on this and thought I was using the KIA (simple) instead of the Benz (edge vector) based on this doc statement:
"
The simple algorithm emulates Blue Iris version 4, but the newer and somewhat more
sophisticated edge vector algorithm is now the default for version 5."

Edit: OK now that you highlighted simple vs. vector, I see in forum multiple posts about random issues with Edge Vector. (depending on circumstances) So, I will focus on that ..Thanks again.
 
Last edited:

pozzello

Known around here
Joined
Oct 7, 2015
Messages
2,270
Reaction score
1,117
you might also want to bump up the 'receive buffer' to 20mb (from default 6)...
 

wittaj

IPCT Contributor
Joined
Apr 28, 2019
Messages
24,436
Reaction score
47,561
Location
USA
In addition, turn off object detection box - that is used to try to knock out false triggers and you have the opposite going on. Most of my cameras have the object detection box unchecked.

Since you are recording continuous, drop the prebuffer down to like 1 second and see if that will improve photos being sent to AI.

What is your CPU usage?

Have you optimized BI per every item in the wiki? Many times people change settings they don't know what they do, and ends up increasing CPU usage and causing other problems:

 

goldbond

n3wb
Joined
Jul 16, 2020
Messages
23
Reaction score
14
Location
Atlanta
In addition, turn off object detection box - that is used to try to knock out false triggers and you have the opposite going on. Most of my cameras have the object detection box unchecked.

Since you are recording continuous, drop the prebuffer down to like 1 second and see if that will improve photos being sent to AI.

What is your CPU usage?

Have you optimized BI per every item in the wiki? Many times people change settings they don't know what they do, and ends up increasing CPU usage and causing other problems:

CPU usage is under 5% until AI detection goes off and then it still stays under 15%.
 

Attachments

wittaj

IPCT Contributor
Joined
Apr 28, 2019
Messages
24,436
Reaction score
47,561
Location
USA
OK, I would turn off object detection as that is now probably the biggest thing preventing some of your triggers - between this feature saying an object needs to move so many pixels and edge vector that you switched to simple you should get your motion triggers.
 

goldbond

n3wb
Joined
Jul 16, 2020
Messages
23
Reaction score
14
Location
Atlanta
I seem to be going full circle... Settings change: Objects detect off completely, simple alg, receive buffer 20 mb, CPU still under 5%....

1. Car based on suggestions and it worked. Detection was accurate/centered in alert. Keep in mind, it worked about 80% of time before...need more time.
2. People are being detected with motion fed to AIInput now. However, DS didn't detect clearly obvious human pictures. DS has not been an issue to date. As long as photos were dropped there, it was 99.99% accurate on detection. I would receive alerts as intended.
 

Attachments

goldbond

n3wb
Joined
Jul 16, 2020
Messages
23
Reaction score
14
Location
Atlanta
I seem to be going full circle... Settings change: Objects detect off completely, simple alg, receive buffer 20 mb, CPU still under 5%....

1. Car based on suggestions and it worked. Detection was accurate/centered in alert. Keep in mind, it worked about 80% of time before...need more time.
2. People are being detected with motion fed to AIInput now. However, DS didn't detect clearly obvious human pictures. DS has not been an issue to date. As long as photos were dropped there, it was 99.99% accurate on detection. I would receive alerts as intended.
Vehicle didn't work at night again. Towel thrown. Either a phd is needed for this product or it's poppycock..
 

fenderman

Staff member
Joined
Mar 9, 2014
Messages
36,897
Reaction score
21,250
Vehicle didn't work at night again. Towel thrown. Either a phd is needed for this product or it's poppycock..
Na, just a user who changes settings and doesnt know what they are doing. Delete the camera and start OVER. Surprisingly many thousands of BI users have no problem with this. This is the developers fault though, he failed to list a minimum IQ of 70 under requirements.
 
Last edited:

wittaj

IPCT Contributor
Joined
Apr 28, 2019
Messages
24,436
Reaction score
47,561
Location
USA
Some locations based on the camera used are just inherently going to have difficulty at night, especially what looks like a 2.8 or 4mm fixed lens trying to pick up a vehicle many feet beyond DORI identify numbers off center of the field of view.

Are you running auto settings in the camera? Running infrared? A 4k and 5MP camera on what is probably a 1/3 or 1/2.7 inch sensor needs a ton of light so the image is probably not good enough at that distance.

Post a video of the night AI miss and we can probably point out the issue and if there is anything that can be done.
 

wittaj

IPCT Contributor
Joined
Apr 28, 2019
Messages
24,436
Reaction score
47,561
Location
USA
An 8MP on a 1/2.7" sensor is going to struggle at night. Not many here would recommend that combination. A 2.8mm makes a good overview, but tough to IDENTIFY a stranger much more than 10 feet away and these same DORI identification numbers inherently come into play with AI like DeepStack.

I am not sure if it is missing vehicles on the cul-de-sac or on the driveway or both, but that picture looks washed out - either settings for contrast/brightness are not correct or you have too much WDR on. Any of that will contribute to DeepStack having trouble, especially at night. But even in the daytime those will cause misses at certain times of the day. While that person is clear to us, I think it blends together for DeepStack.

Not having enough contrast (either via settings or WDR) will cause misses at night with infrared.

I would also suggest adjusting your camera angle to not have any of the soffit in the field of view.

How many images do you have this camera sending to DeepStack? Remember, DeepStack needs time to see and recognize the object. I would try upping it to 5 or 10 images and repeat and see if you get some hits. And what confidence level are you running at? At night and with objects further away, you have to drop the confidence level down as well.
 
Top