BI/Deepstack haven't alerted on a car pulling into driveway - need help answering "why?"

wittaj

IPCT Contributor
Joined
Apr 28, 2019
Messages
24,987
Reaction score
48,731
Location
USA
Yep, turn off the object detection features off - a car turning around didn't travel 100 pixels and the reset probably reset the trigger with the headlights.

Object Detection is used to knock out false triggers, and you have the opposite going on.
 
Last edited:

JL-F1

Getting the hang of it
Joined
Jun 12, 2020
Messages
119
Reaction score
73
Location
USA
as wittaj said, use intrusion detection in the camera instead of BI Motion. For me in testing, I have the same camera 5442T-ZE, that is close to 100% for cars in my situation, at least for triggering. I use that with external triggers and then I have BI use DS to analyze. I have IVS set for cars or people, then the camera triggers BI then DS to let me know if it was a car or a person.
 
Joined
Sep 19, 2015
Messages
956
Reaction score
1,217
Location
Naples Fl
as wittaj said, use intrusion detection in the camera instead of BI Motion. For me in testing, I have the same camera 5442T-ZE, that is close to 100% for cars in my situation, at least for triggering. I use that with external triggers and then I have BI use DS to analyze. I have IVS set for cars or people, then the camera triggers BI then DS to let me know if it was a car or a person.
do you have any screen shots you could share?
 

JL-F1

Getting the hang of it
Joined
Jun 12, 2020
Messages
119
Reaction score
73
Location
USA
In the first pic: 'Rule' config' draw a rectangle around the area you want to check if a car enters.

The BI DS settings are same as usual
 

Attachments

Last edited:

mad-maks

n3wb
Joined
Oct 18, 2020
Messages
16
Reaction score
3
Location
New York
Alright, changed BI settings (unchecked "object travels"). Going to leave it like this for several days.

I will probably play with camera's intrusion detection as well
 

sebastiantombs

Known around here
Joined
Dec 28, 2019
Messages
11,511
Reaction score
27,695
Location
New Jersey
I use both, usually clone a camera to use ONVIF in one and BI motion in the other. That way nothing gets missed and it's easier to see which one actually tripped.

Keep us posted how things are working out.
 

wittaj

IPCT Contributor
Joined
Apr 28, 2019
Messages
24,987
Reaction score
48,731
Location
USA
Unchecking this will keep "Detect/Ignore Static Objects" from working according to Ken.
Wow that is interesting. None of my cameras use that object travels option and I do not get any false DS alerts due to static objects if the camera is triggered by a bird....
 

105437

BIT Beta Team
Joined
Jun 8, 2015
Messages
2,046
Reaction score
951
Wow that is interesting. None of my cameras use that object travels option and I do not get any false DS alerts due to static objects if the camera is triggered by a bird....
Yeah, he said it on the last webinar and I was surprised as well.
 

sebastiantombs

Known around here
Joined
Dec 28, 2019
Messages
11,511
Reaction score
27,695
Location
New Jersey
Now I know why DS is detecting static objects for me. Time to play with that option. DS/BI is turning into a fulltime job.
 

wittaj

IPCT Contributor
Joined
Apr 28, 2019
Messages
24,987
Reaction score
48,731
Location
USA
I wonder if that was changed with a more recent update?

I was on the bleeding edge early on when DS came out and I try the newer versions, but the custom models and/or face bomb out and cause all sorts of issues, so I am on the update just before custom models was added as that is my most stable version and do not get it detecting static objects. They pop up during tuning, but not live.
 

sebastiantombs

Known around here
Joined
Dec 28, 2019
Messages
11,511
Reaction score
27,695
Location
New Jersey
I haven't had much luck with custom models either. Neither logo or dark work worth a darn and add significant time to analysis from what I see. I am on the bleeding edge and the built-in stuff works fine, during the day and with slow moving things at night.
 

kklee

Pulling my weight
Joined
May 9, 2020
Messages
187
Reaction score
203
Location
Vancouver, BC
Just adding my experience with the DS analysis tuning over the last couple of weeks.

The custom dark model seems to be working well for me. It's picking up people and cars better at night. During the day, it will often show a higher confidence for a person than the built-in model.

Faces has been very hit and miss. It sometimes works and flags an unknown (about 25%), which intermittently captures a face image (maybe only about 10% of the time). Granted, my cameras are not configured for optimal face capture.

Faces also seems to interfere with object analysis and causes intermittent crashes of BI. With faces enabled, DS will occasionally miss detecting people. Running the clip that did not trigger an alert through analysis will show that DS does detect the person, but the alert was cancelled as nothing detected.

I have object detection turned on for all cameras, I find that cuts down on false alerts due to shadows and flora waving around. The static object detection works intermittently. The barbeque on my deck is picked up as a person quite often. I don't know if the faces option was interfering with that, I just turned faces off and will need to monitor for a bit.

I played with the "Apply to motion triggers only" setting and found that DS would miss detections more often with it enabled. Again, this might have been due to faces being enabled, will need to play with it some more.
 

wittaj

IPCT Contributor
Joined
Apr 28, 2019
Messages
24,987
Reaction score
48,731
Location
USA
I haven't had much luck with custom models either. Neither logo or dark work worth a darn and add significant time to analysis from what I see. I am on the bleeding edge and the built-in stuff works fine, during the day and with slow moving things at night.
I have bounced back and forth between bleeding edge and 5.6.4.3 (the one right before the custom models was added).

I have noticed that DS time is 3 times longer than it was when it first started, even rolling back to the prior version.

Can you check your logs on times and see if your DS time is longer after trying a custom model and then rolling back.

Now what is interesting is when I went bleeding edge, I added in the models file to use box face:0, dark so that it would skip face model and run the dark when I was trying to get rid of the server error 100. When I rollback that option is obviously not there, but when I try bleeding edge, those inputted items return, so BI is keeping that even if you rollback and I wonder if rolling back is still using some things that is slowing it down now?
 

sebastiantombs

Known around here
Joined
Dec 28, 2019
Messages
11,511
Reaction score
27,695
Location
New Jersey
If I remember the numbers correctly, detection times almost doubled when adding a model, or two. My machine is kind of slow in that department and I normally see times in the 300-400ms range. With custom models it would go as high as 1200ms.
 

wittaj

IPCT Contributor
Joined
Apr 28, 2019
Messages
24,987
Reaction score
48,731
Location
USA
If I remember the numbers correctly, detection times almost doubled when adding a model, or two. My machine is kind of slow in that department and I normally see times in the 300-400ms range. With custom models it would go as high as 1200ms.
So yours went back down when you stopped the custom models?
 
Top