Night and DeepStack camera settings

Charles_CO

Young grasshopper
May 4, 2022
35
14
Denver
Is there anything special I need to do to improve AI detection at night? I have pretty bad luck with my setup when it comes to night detection.

Bellow is screenshot to illustrates the situation that resulted in cancellation because nothing was found. In another instance I was recognized as a dog. I assume something must be wrong with my settings. I could not really find profile settings on my camera like some of you have. Like setting profile for a day and night.
That said I don't think the picture quality it terrible. I understand that the camera is pretty high but that should not be a big problem as during the day it detects person just fine. I have People and Person to confirm for night AI profile as I was not sure which one applies for dark model now.

Here is link to video from the camera



Any suggestions or tip on improving the situation?

Thank you. Charles
 

Attachments

  • setting.jpg
    setting.jpg
    424.1 KB · Views: 37
  • Night_1.jpg
    Night_1.jpg
    334.6 KB · Views: 36
Nighttime tends to be problematic, especially B&W images.

Camera is a little high as you said.

You need to adjust brightness/contrast/etc. so that there is even more pronounced differences. To the camera, you kinda blend in with the hotspot that makes everything white and then a shirt with vertical lines that mimic the decking. Now I realize that it will happen with other clothing as well, but the point is the camera needs to see enough difference in order for Deepstack to recognize it.

Or try running in color if you have enough light.
 
  • Like
Reactions: sebastiantombs
I guess I am out of luck without profiles on the camera. If I adjust things to the extreme at night it will look like s**t during the day. I will try to play with settings a little bit.
 
 
What exact camera?
2x GW Security - GW8555MIP
and
1x Camius BoltV

I bought them way back and I knew nothing back then about cameras.
User interface looks pretty much same for both so I am not sure if it is same manufacturer or just software that is used on both. I don't see profile option on any of them.
 
How about a screen shot of the cameras tab of the console showing frame rates, bit rates, and key frame (iframe) rates.
 
@sebastiantombs , I haven't really messed with settings yet but it looks like I will have to find some compromise for both day and night. I have 3 Dauha cameras coming in soon to include in the mix.
 

Attachments

  • 3.jpg
    3.jpg
    179.2 KB · Views: 25
  • 2.jpg
    2.jpg
    238.7 KB · Views: 25
  • 1.jpg
    1.jpg
    174.3 KB · Views: 22
Last edited:
first observation is to make the frame and iframe rates the same. Blue iris likes it that way for the best motion detection. 15F/ps is plenty for surveillance cameras, 20 and 30 is not "buying" anything in terms of functionality so I'd slow them down to 15F/ps.

I was asking for a screen shot from the Blue Iris console, not the cameras. In the console click on the lightning bolt/graph icon on the upper left. Then click on the cameras tab and post a screen shot of that preferably at night.
 
  • Like
Reactions: looney2ns
Oh.. I misunderstood..
I kept highest settings for main stream because I like to mess with videos when I catch something nice..
Maybe it's not best for general surveillance. The screen shot is below the video links.


That's from old house:
Bears


Mountain Lion


Messing with a deer - cameras and home automation in action. LOL
 

Attachments

  • 10.jpg
    10.jpg
    348 KB · Views: 23
From that screen shot of the camera stats some of them are running on the low side for a key frame (iframe) rate. I can't be sure which cameras they are in relation to your initial problem, but that's something else to address.

15F/ps or 30F/ps if the shutter speed is correct you'll still get good captures. Hollywood uses 24F/ps for a big screen movie.

Back to the original problem. It all comes down to contrast for DeepStack to be able to "see", especially at night. Try working with brightness and contrast as was suggested. I have a camera that is a real problem chld that way and I'm constantly tinkering with it trying to get better performance out of it in terms of DeepStack detection.