Deep Stack is not helping.

[GaryK4]

My main camera has a tree in the yard. I get many triggers when the wind blows. I install Deep Stack some time ago and it does not seem to reduce the tree trigger. Am I doing something wrong?
Motion sensor checked. object size 126, contrast 20, min duration 1.0, Algorithm: Simple
When triggered: Database only, Immediate actions enabled but no items added.
Artificial Intelligence: DeepStack, Override Server set, All checkboxes checked except: Burn label, Auto-flag and Plate Recognizer.
To confirm: person,car,truck,bus,bicycle,boat To cancel and Custom not checked.
min confidence: 80%, images 2, analyze 750ms

 

[GaryK4]

I am very disappointed that no one has and tips or feedback on this issue. 90% of my recording is of wind blowing.

 

[Swampledge]

It might help if you provide more information about what’s happening. Do you even know if images are being sent to DS?

 

[Swampledge]

Also, terminology is important. Triggers are caused by motion detection. Alerts , if DS is functioning for you, should only occur after confirmation. If you don’t have cancelled alerts, DS isn’t analyzing your motion trigger images.

 

[bp2008]

Yes, to be clear, AI processing is too slow so it does not prevent recording. It can however prevent an "alert" being created.

 

[wittaj]

Screenshots of all relevant pages are important, as is a screenshot of the field of view.

The all text above means nothing. As an example Object size 126 and min duration of 1.0 second - depending on the field of view that may or may not be appropriate. That would miss every plate on my LPR camera as an example.

 

[GaryK4]

Also, terminology is important. Triggers are caused by motion detection. Alerts , if DS is functioning for you, should only occur after confirmation. If you don’t have cancelled alerts, DS isn’t analyzing your motion trigger images.

OK, I am not getting any alert images for this camera. Other camara's do have alert images. On the main camera, I have Save DeepStack analysis details set, but I don't know where that data is.

 

[GaryK4]

Screenshots of all relevant pages are important, as is a screenshot of the field of view.

The all text above means nothing. As an example Object size 126 and min duration of 1.0 second - depending on the field of view that may or may not be appropriate. That would miss every plate on my LPR camera as an example.

I am not trying to get plates at this time, just minimize shadow recordings from tree movement. Zones don't work because time of day changes the shadow location too much.

 

[Swampledge]

You can view DeepStack analysis details by doing a Ctrl-Double Click on the alert image. (Which you don’t have.). As wittaj suggested, we can best help if you post images of your trigger, motion detection, etc. configuration screens. Do you have any cancelled alerts?

 

[Swampledge]

Also, you can read through this thread to see if it helps you any:

Deepstack confirming alerts despite zone exclusions

Hi there - I've tried to review threads and read the BI help docs carefully, but I think I must still be missing something essential about how the various settings work. I have a camera pointed out my driveway, where our two cars park. It also includes the road in front of our house in the...

ipcamtalk.com

 

[GaryK4]

Yes, to be clear, AI processing is too slow so it does not prevent recording. It can however prevent an "alert" being created.

This may well be the issue as I am not getting alerts for this camera.

 

[wittaj]

I am not trying to get plates at this time, just minimize shadow recordings from tree movement. Zones don't work because time of day changes the shadow location too much.

I used plates an an example in my case. But FYI - you won't get plates at that far out. Maybe on the driveway, but not the street.

So that is a wide angle view, you could make the min time longer than 1 second. Only a superhero would be in and out of the field of view in under 1 second LOL. Try 2 seconds and then walk around and drive the street and confirm they will trigger.

 

[GaryK4]

You can view DeepStack analysis details by doing a Ctrl-Double Click on the alert image. (Which you don’t have.). As wittaj suggested, we can best help if you post images of your trigger, motion detection, etc. configuration screens. Do you have any cancelled alerts?

 

[wittaj]

That is a pretty small min size for that field of view. Any reason why you have it so small?

Uncheck the use mainstream as AI will downrez it anyway, so using mainstream is only wasting CPU and time.

With it that small, every tiny leaf movement will trigger and then you are only doing 2 images, so it is very possible that the leaf blew and then someone walked in and it misses it.

You probably need to add more photos to be analyzed and add something in the cancel box like giraffe to force it to analyze all the images.

I think you could still use zones. At that small object size, the leaves way down the street will trigger the cam.

 

[Swampledge]

@GaryK4, What does your Alerts config tab look like?

 

[GaryK4]

That is a pretty small min size for that field of view. Any reason why you have it so small?
Because I was not getting all the motion that was "real"

Uncheck the use mainstream as AI will downrez it anyway, so using mainstream is only wasting CPU and time.

With it that small, every tiny leaf movement will trigger and then you are only doing 2 images, so it is very possible that the leaf blew and then someone walked in and it misses it.

You probably need to add more photos to be analyzed and add something in the cancel box like giraffe to force it to analyze all the images.

I think you could still use zones. At that small object size, the leaves way down the street will trigger the cam.

OK, I will try some of these suggestions.
Thanks

 

[GaryK4]

@GaryK4, What does your Alerts config tab look like?

 

[wittaj]

It might also be the case of trying to do too much with one field of view.

Keep in mind what AI is - it is someone took photos from cameras and identified/labeled items in the photo and then your image is being compared to them. So a person way out on the edge of your field of view may be too small for AI to identify it.

But looking at your image here, if you really wanted it to trigger for that person way down the road to have DeepStack start analyzing, then make the minimum size the person at the end of the SUV.

But again, that might be too small for the model that is being used. Realistically you shouldn't be trying to use it for anything beyond the red box, so you have someone stand next to your mailbox and draw the min size around that person. And proper uses of zones can help knock out the tree triggers as well.

What camera is this? Maybe it has AI in it and you would be better off using it instead of Deepstack?

 

[GaryK4]

It might also be the case of trying to do too much with one field of view.

Keep in mind what AI is - it is someone took photos from cameras and identified/labeled items in the photo and then your image is being compared to them. So a person way out on the edge of your field of view may be too small for AI to identify it.

But looking at your image here, if you really wanted it to trigger for that person way down the road to have DeepStack start analyzing, then make the minimum size the person at the end of the SUV.

But again, that might be too small for the model that is being used. Realistically you shouldn't be trying to use it for anything beyond the red box, so you have someone stand next to your mailbox and draw the min size around that person. And proper uses of zones can help knock out the tree triggers as well.

What camera is this? Maybe it has AI in it and you would be better off using it instead of Deepstack?

View attachment 166851

Dahua DH-SD22A204TN-GN PTZ camera.
I don't think it has motion detection. If it did, can that trigger be passed to BI for recording? It may be worth upgrading.

 

[wittaj]

Yeah that is a lite series camera so it will not have AI and its motion detection would be worse.

Upgrade to a 5442 and the AI trigger can be sent to BI and you can get rid of all that Deepstack stuff since your use case is vehicle and people.

Who uses Dahua AI capable cameras? Reliable AI for triggering events? Pro's/con's?

It has come to my attention that certain Dahua cameras come with built in AI that distinguishes between human & cars. I was unaware of this. Well, always saw the AI word used in the reviews but never really thought about it because I thought it was a Dahua NVR integration only. I am currently...

ipcamtalk.com