DS & Motion Detection Rectangles

Shap

Shap

Young grasshopper
Jul 4, 2021
37
22
Austin,TX
Hello,

I have a problem with my BI+DS configuration and need some help.
Basically, in many cases, I get something like this: (any ideas why motion rectangles are not synced with the actual object?)
I have Object Detection turned on.

1628598922149.png
 
Looks like a processing speed issue wrt Deepstack. Do some experimenting...reduce DS mode to Low, don't use any custom models including facial recognition, try reducing your interval to 500msec in the cam's AI tab and set your make time (trigger tab) to less than a second. If you're using the CPU version of DS, consider trying the GPU version if you have an Nvidia card and the wherewithal to set it up. You can also experiment with a short pre-trigger buffer...maybe 3 seconds or so. See what some of those suggestions do for you--good luck!
 
  • Like
Reactions: sebastiantombs
Thank you - I did. The problem is that even when using the substream, I still get the following problems:

1. There is a mismatch between object location and motion detection causing DS to miss alert
2. I still can not explain why there is a 2-sec delay between the first 2 images

Here is the image from using motion overlay from direct-to-disk:
1629213016461.png
 
Shap said:
Thank you - I did. The problem is that even when using the substream, I still get the following problems:

1. There is a mismatch between object location and motion detection causing DS to miss alert
2. I still can not explain why there is a 2-sec delay between the first 2 images

Here is the image from using motion overlay from direct-to-disk:
View attachment 98761
Click to expand...
I had the same issue. Ken suggested using the substream for DS. Camera Properties > Trigger > Artificial Intelligence > Unselect "Üse main stream if available". Resolved the issue in my case....
 
Yep, I am already using the substream. I try to understand why there is a 3-4 sec delay between T-x and T+x.
 
Also, pretty often, DS will miss some events as the motion coordinates are not sent correctly.
For example, when I select a clip and do Ctrl + mouse double click (that should bring me to the motion detection frame), I will get:
(you can see that the motion frame that was passed to the DS is a different one, and the motion rec

From all the debugging I did, it seems there is a problem with frames that BI send to DS and the motion location.
The is no way to really debug this as BI or DS do not provide an ability to save aside the image that was sent for analisys
and motion coordinates.

1629809917466.png
 
What is the red bar across the bottom and next to all of your clips? I must've turned that setting off. If that's system resources, it looks pretty slammed.
 
stev067 said:
What is the red bar across the bottom and next to all of your clips? I must've turned that setting off. If that's system resources, it looks pretty slammed.
Click to expand...

It is just the color I selected for the camera. :)
 
  • Like
Reactions: stev067
YrbkMgr said:
My observation as well. It doesn't happen all of the time, but enough to conclude that there's an issue.



Brilliant tip !!! I had no idea that existed - that makes things a lot easier for troubleshooting. How did you find out about that? Tribal knowledge or is it documented somewhere?



Definitely, but not always, right? Sometimes it behaves as expected - so I've been trying to analyze the dat information for successful confirmed alerts first - is there anything in common that I can use to evaluate the dat files of the falsely cancelled alerts... kind of approach. Since Ken organized BI settings into logical groups instead of functional ones though, finding out what settings to "guess and check" i.e., what settings impact the results, is maddeningly frustrating.



Well, the image(s) used are captured in the dat file. And the "decision" image is the one with the asterisk. Each timestamp has a unique image and the number of images are driven by the AI settings in the Trigger tab. Clicking on each timestamp shows the motion zones DS was using at that moment. So the images used in the analysis are stored within the dat file.
Click to expand...

I think somebody sent me a link for this tip. I do not have this link right now, but I am sure it was from Blue Iris - Index page
 
  • Like
Reactions: YrbkMgr
Good point. The only problem it is not sending this frame to DS :(
 
jaydeel said:
Just confirming ... when you Ctrl-double-click the alert in the cliplist, the Blue Iris playback window is not showing the *.dat file frame marked with an asterisk?
View attachment 99628
Click to expand...

Not exactly - here what happens when I click Ctrl+Double Click: (frame on the left is not in the list on DS events on the right - none of them will show motion in the same place as left frame).
1629912142602.png

BTW, another issue that is happening 100% of the time - see that the first two events sent to DS as exactly the same and processed twice (I am sure they were sent twice as you can see processing time by DS is different):
(look at T+1793 msec)

1629914533460.png
 
There is an
jaydeel said:
I wonder if this could be a coding artifact and for some reason the T+1793 msec JSON string is rendered twice to the graphic display.

My only suggestion is to email the screenshot and *.dat file to support@blueirissoftware.com.
Click to expand...
Already did - the problem with the support is they are SO SLOW on response. Weeks.

Why they do not have GitHub or something similar is beyond my understanding ...

Guys - anybody else has this problem (same T+X) sent twice to the DS?
 
Shap said:
There is an

Already did - the problem with the support is they are SO SLOW on response. Weeks.

Why they do not have GitHub or something similar is beyond my understanding ...

Guys - anybody else has this problem (same T+X) sent twice to the DS?
Click to expand...
I think they probably try to prioritize their issues. The double frame doesn't seem like that big of a deal, while it may be a bug. Just add an extra to the end. But do the json details of the two matching frames also match?
 
^^ I concur with checking the JSON details for both frames.

Blue Iris is one programmer (Ken) and a few support staff. We've all had to learn to be patient. The product is worth the wait. :)
Personally, I'm hoping they're taking a bit of vacation.
 
  • Like
Reactions: slabbel
jaydeel said:
^^ I concur with checking the JSON details for both frames.

Blue Iris is one programmer (Ken) and a few support staff. We've all had to learn to be patient. The product is worth the wait. :)
Personally, I'm hoping they're taking a bit of vacation.
Click to expand...

I did not know that - that explains the delays. Not a problem in my case, I am doing this mostly for fun :)
So what I did now I removed BI, DS, cleaned the registry. Reinstalled.
I have only 3 cameras. Connected them (not from backup).

The problem is almost solved :
1. No more double frames
2. No more significant delays for the first motion frame
3. I still do not get T+0 - now it is between T+60 and T+100 msec. Not a big deal, but I do see that the frame that was sent to DS is a bit off from the motion rectangle position.

I do not know what caused the issue, I will start adding my configuration again (MQTT, email notifications, etc) and will see if the problem will be back.
The problem caused by this setting - immediately after setting it to hi-res jpeg problem reappeared:
1629945581207.png


P.S. The problem in IMHO is the absence of a detailed debug log.
 
Last edited:
  • Like
Reactions: YrbkMgr
You must log in or register to reply here.
Top Bottom