Night Detection

OSU_BuckeyesFan

Young grasshopper
Apr 30, 2023
43
10
Ohio
Using Blue Iris and CodeProject usually works well during the day for person detection but I struggle with night/dark detection, I'm assuming this is more of a camera problem than an AI problem but I could be wrong. I'm looking for advice or suggestions as to what I can do to clean up this image for better detection. The attachment is a screenshot of a video from this morning when I was leaving for work, motion was detected by the camera however AI didn't pick me up and determine that I'm a person. I can change the exposure on the camera to make the image brighter but that doesn't seem to help, it just makes the person blurry like a ghost. Within Blue Iris, I'm using ipcam-general as the model which I thought would be best for day/night person detection, should I be using a different model instead? Any other recommendations or suggestions?
Test Pic.jpg
 
  • Like
Reactions: mat200
Nighttime always brings the struggles.

Dark grass, dark clothing, blur motion. The AI wasn't designed for that so it can struggle.

Add more light or try infrared where you can speed the shutter up so it isn't such a blur or get a better camera.
 
This camera is a Hanwha QNV-7010R.
That camera is on a poor image sensor, thus it will struggle at night, as you can see in the blur of you moving.
It's also too wide of view as a 2.8mm lens.
Plus the dome format is problematic all around.

Best all around camera at this point is THIS
The performance of this cam will be head and shoulders better than your current cam.

 
That camera is on a poor image sensor, thus it will struggle at night, as you can see in the blur of you moving.
It's also too wide of view as a 2.8mm lens.
Plus the dome format is problematic all around.

Best all around camera at this point is THIS
The performance of this cam will be head and shoulders better than your current cam.

Thanks for the reply wittaj and looney2ns.
-Why do most of the larger and more well-known manufacturers such as Hanwha, Axis, & Hikvision have most of their cameras with sensors of 1/2.8 if it's not ideal?
-A 2.8mm lens gives me a good overall view of my driveway area but I do understand it is too wide for more detail. What lens size would be the max recommended for this area?
 
Thanks for the reply wittaj and looney2ns.
-Why do most of the larger and more well-known manufacturers such as Hanwha, Axis, & Hikvision have most of their cameras with sensors of 1/2.8 if it's not ideal?
-A 2.8mm lens gives me a good overall view of my driveway area but I do understand it is too wide for more detail. What lens size would be the max recommended for this area?

Because they are cheap, allow to use smaller and cheaper optics / lenses and can be put in smaller / cheaper chassis...

Also this allows to have higher models (PRO or ULTRA lines) with bigger sensors which can be much much pricier..

PS. Picture of moving car at night in Color with ultra line camera (Dahua 7442-Z4) with big 1/1.8" sensor and very good lenses..

moving.jpg

Different installation with Dahua 5442-Z4E (lower line, also big 1/1.8" sensor) and lenses with strong zoom...
All cars snapshots, car descriptions and car plates snapshots done by AI on camera...

VMD vehicles night.png
 
Thanks for the reply wittaj and looney2ns.
-Why do most of the larger and more well-known manufacturers such as Hanwha, Axis, & Hikvision have most of their cameras with sensors of 1/2.8 if it's not ideal?
-A 2.8mm lens gives me a good overall view of my driveway area but I do understand it is too wide for more detail. What lens size would be the max recommended for this area?
I detect some confusion.....

1/2.8" is sensor size, in inches. It is suitable for a 2MP camera with minimal ambient lighting.
A 2.8mm lens is the focal length, in millmeters.

Ideal-sensor-size-to-megapixel.png
 
Last edited:
  • Like
Reactions: JDreaming
might be the camera.
the 5442 i have on CP ai isnt on a driveway, and I'm not seeing anybody out at night, but look how far away it decided to flag " person" on this clip.
1730448253226.png
 
  • Like
Reactions: JDreaming
Maybe try with NO code Project?
at dusk in full color mode the 5442 totally flopped on catching these trick or treaters with CP ai at 6:41. maybe i could set a lower %age of ( I switch from 50% to 35% tonight)
but i need 24 hours to see how it does now.
Thee olde 3241ZAS- 2MP on a 1/2.8 sensor with NON Ai detected the kids in motion.
If I ran IR on the 5442 it would have had enough contrast to " see " them. But I'm more interested in the color of the cars going by at night.
1730448652548.png

1730448727807.png
 
Last edited:
  • Like
Reactions: JDreaming