Deepstack motion detection tuning...

[1fxman]

Ugh... fakk. Totally get it. But can't you view that link without logging in? That links directly to the deepstack bit

Yes. I just read the first part. My problem is that I can’t get in front of the BI computer in order to try out what the forum suggests. If I can ever figure out how to use UI3 i could mess with it remotely from my house. Thanks

 

[wittaj]

Are you port-forwarding or OpenVPN to review it remotely? Or none, in which case you won't review remotely...

 

[dan2112]

Does anybody have recommendations on fast moving objects, like speeding cars. I have tuned the number of real time images down to zero and have 750 for the analyze one each. But the cars are at night and there is a lot of headlight glare. The logs show no deepstack activity at that time and when I replay it through deepstack I am lucky to get one recognition event. Suggestions?

 

[1fxman]

Are you port-forwarding or OpenVPN to review it remotely? Or none, in which case you won't review remotely...

Port forwarding

 

[dan2112]

Are you looking to be alerted on fast moving objects, or are you getting false alerts on fast moving objects?

For the sake of clarity, the camera is triggered by motion sensor settings. If the camera enters a triggered state, recording occurs (assuming not continuous). Separately, deepstack evaluates that motion and determines whether or not to alert you. This post might be worth checking out.

I am not getting any alerts. Only reviewing the video do I see the car. The paper delivery guy is notorious for ripping up and down the streets at 4am.
Maybe I should turn down the min duration?

 

[1fxman]

I figured out why I couldn’t use UI3 remotely. I was leaving out a caricature in my password.

 

[sebastiantombs]

You need to add some "real time" images since that's what DS uses for analysis. Having it set at zero is stopping it from analyzing anything. I'd suggest starting with 5 and it may need to go higher depending on many variables.

If the target is fast moving, and that is relative depending on view and target speed, I'd try lowering the "analyze one every" xxx seconds to 500 or even 250 milliseconds, IE every half or quarter of a second.

Edit - I should have mentioned that pre-trigger time will have a direct effect on how many images are needed for reliable detection. If it's on the long side, say greater than two seconds, you'd need 8 images to get to the point where the target is in the frame at 500ms. So there it would need to be at least 12 or higher depending on contrast, brightness, speed and so on.

 

[wittaj]

And at 4am, it is dark, so if you are trying to do too much with your field of view, DeepStack may miss the object altogether.