Bad quality for snapshot day/night

[Corby59]

Hi everyone,

i have a DH-SD1A404XB-GNR camera, and activated face detection, but the quality of the snapshots is really disgusting whether during the day or at night.
In your opinion, what are the settings to have a good image quality for snapshots?
Here are my current settings, I have done tests with different settings, the quality is always bad.






Thanks in advance

 

[bigredfish]

Bad when viewed where?

Are you seeing the low res placeholder shots on the NVR AI search pages? Please give example

Also on your main recording, Use
h.264.H
CBR
Match Iframe to FPS (so both 25)
if 4MP cameras increase bitrate to at least 10240

 

[Corby59]

Bad when viewed where?

Are you seeing the low res placeholder shots on the NVR AI search pages? Please give example

Also on your main recording, Use
h.264.H
CBR
Match Iframe to FPS (so both 25)
if 4MP cameras increase bitrate to at least 10240

Hi Bigredfish (fisherman ?)

I mainly view captures directly via the email notifications I receive when a face detection occurs.
Sorry, but I don't understand what you mean by your 2nd question.
All detections videos/snapshots go to my Synology NAS.

 

[bigredfish]

Those are typically low res versions.
Show an example pic directly from an email notification...

VBR may well be your biggest problem

 

[steve1225]

I mainly view captures directly via the email notifications I receive when a face detection occurs.
Sorry, but I don't understand what you mean by your 2nd question.
All detections videos/snapshots go to my Synology NAS.

He is talking about face snapshots (1 inch cutoffs - like in Video MetaData mode) not video recording quality...

Hi everyone,

i have a DH-SD1A404XB-GNR camera, and activated face detection, but the quality of the snapshots is really disgusting whether during the day or at night.
In your opinion, what are the settings to have a good image quality for snapshots?
Here are my current settings, I have done tests with different settings, the quality is always bad.

Show us examples...

Probably reason:
This is cheap mini PTZ cams without human auto tracking...
You have very wide view (You don't use optical zoom much)..

In this case humans and faces are small in full video.
Face snapshots are cut from video - camera can't increase resolution of them...

So You have small face snapshots..
Zoom (optically) more camera to area where people are walking (street).

Then You will got bigger humans & faces on videos => better faces snapshots in email...

ps. I don't know is this camera model taking face snapshots from video feed or big image snapshots.. if second option, then You should increase resolution of big snapshots to full camera resolution (4MP)..

 

[Corby59]

Those are topically low res versions.
Show an example pic directly from an email notification...

VBR may well be your biggest problem

Here, a pic with old parameters (VBR):


And here, with the your parameters (CBR):

 

[TonyR]

FWIW, bottom image above is only a thumbnail-sized 168 x 200; snapshot setting is set to 1920 x 1080 so it seems the thumbnail image is created from a still of the video feed.

 

[Corby59]

So if I understand correctly, I won't be able to have a better quality for the snapshots, is that right?

@steve1225:
The problem is that if I zoom in on the detection area, I don't have enough distance and therefore the field of vision will be reduced.

 

[steve1225]

So if I understand correctly, I won't be able to have a better quality for the snapshots, is that right?

@steve1225:
The problem is that if I zoom in on the detection area, I don't have enough distance and therefore the field of vision will be reduced.

Yes... You must decide - do you want overview camera (which will have small people/faces) or zoomed one which gives You big snapshots...
Normally in situations like this there should be 2 cams - one overview, one zoomed..

Here 2 5442 - one overview 3.6mm (90 degrees horizontal), one 5442-Z4E zoomed at street (it's not max of optical zoom here, owner wanted to see people/cars next to plot / entrance)..



Second zoomed camera gives snapshots like this:

 

[Corby59]

@steve1225

In fact, on this camera, I use face detection because I'm tired of false alerts about my neighbor's cats that keep passing by....
Because even if I activate smart motion detection (human only), I still have false alerts because of cats, or car headlights.

 

[bigredfish]

You should use IVS and not MD/SMD you will get fewer false alerts

 

[steve1225]

@steve1225

In fact, on this camera, I use face detection because I'm tired of false alerts about my neighbor's cats that keep passing by....
Because even if I activate smart motion detection (human only), I still have false alerts because of cats, or car headlights.

Face Detection IS NOT for perimeter protection.
Face Detection works only is face is visible (someone is watching into camera side). This will not catch people from back..

If you look at my demos above at Human detection screen, you will see that some records have also Face snapshots but most not. Because someone was walking in way that face wasn't visible. In that case You will NOT have notifications if You use only Face detect.

You should use IVS. Add tripwire rule, draw zigzag over area where You want catch people/cars..
In IVS you can configure email notifications, snapshots will have full human image (not only face) in any position (front/back/side)...

Depending how many zigzag lines You will draw, then You will have more of notifications with more pictures (they are from point when human cross zigzag line)...

 

[Corby59]

You should use IVS and not MD/SMD you will get fewer false alerts

I tried for a while, but the problem is that the cats keep triggering the camera multiple times a day including at night.

 

[steve1225]

1. in IVS when editing rule you can draw zigzag little higher (to no catch small animals on floor).
2. In IVS when editing rule You can use size function and draw minimal size of someone (bigger that cat), which will create event..

 

[Corby59]

1. in IVS when editing rule you can draw zigzag little higher (to no catch small animals on floor).
2. In IVS when editing rule You can use size function and draw minimal size of someone (bigger that cat), which will create event..

For point #1:
It is difficult for me to draw zigzags a little higher because I risk taking people who pass on the sidewalk, and I don't want that.
What interests me is to take only the people who can pass in the 1st zone:


For the second point:
I have to wait for a cat to pass by to calculate the minimum pixel size.

 

[steve1225]

This is not a zigzag.

Are You using tripwire or intrusion?
For intrusion this is ok - you drew area for intrusion..
for tripwire You should draw zigzag - which will catch people at crossing points.. Something like...

 

[Corby59]

This is not a zigzag.

Are You using tripwire or intrusion?
For intrusion this is ok - you drew area for intrusion..
for tripwire You should draw zigzag - which will catch people at crossing points.. Something like...

it's not a IVS rule, it was just to show (with paint) the area i need to detect intrusion human.
This is the tripwire i have :

​

 

[steve1225]

Tooo muuuuch

 

[Corby59]

Hi everybody,

Do you know why the "Target Filter" box is grayed out?
I no longer have access to this box, even when I deactivate the rule.
I even did a reset to default, and this box is still grayed out.

On live view, i don't see the tripwire:


On my overs Dahua camera, i see the tripwire on live view, and the "target filter" box is not grayed, i can check or uncheck the box.

Thanks.

 

[wittaj]

The cameras with AI do not allow you to uncheck the Target Filter box completely.

It has been an issue for years with people that wanted to turn that off to track animals and to track simply on min object size.