Face Detection
- Thread starter JFire
- Start date
Fastb
Known around here
Jfire,
Not completely.
In Preview, I can see it track faces.
I haven't tried snaps or alarms yet.
Not completely.
In Preview, I can see it track faces.
I haven't tried snaps or alarms yet.
How are you setting up the boxes? On playback is it capturing faces? I've tried setting it up but I don't see anything through the nvr. I'll try through camera this week.
Sent from my SM-N920T using Tapatalk
Fastb
Known around here
JFire,
I'll check Wed.
From memory, playback wasn't real clear on tripwire/intrusion/face detect unless you watch playback in 1X mode. Which I don't do often, while bench testing my system setup.
More later
I'll check Wed.
From memory, playback wasn't real clear on tripwire/intrusion/face detect unless you watch playback in 1X mode. Which I don't do often, while bench testing my system setup.
More later
If you look on playback there's a button of a persons silhouette on the right, I believe that will show all the face captures.
Sent from my SM-N920T using Tapatalk
Fastb
Known around here
@JFire,
Thanks, I hadn't noticed the person silhoutte before.
In my case, it's greyed out. Even watching playback with faces, and FD enabled.
At the bottom of the playback screen, below the colored bars, is a legend:
Do you see orange in your timeline for a FD event?
I was seeing orange with tripwire and/or intrusion zone events.
Thanks, I hadn't noticed the person silhoutte before.
In my case, it's greyed out. Even watching playback with faces, and FD enabled.
At the bottom of the playback screen, below the colored bars, is a legend:
Do you see orange in your timeline for a FD event?
I was seeing orange with tripwire and/or intrusion zone events.
Last edited by a moderator:
@Fastbi haven't gotten face detection to work at all
Sent from my SM-N920T using Tapatalk
Sent from my SM-N920T using Tapatalk
Last edited by a moderator:
Fastb
Known around here
@JFire,
Sorry to hear that.
I had it working, ever so briefly. Below is an example of an email I received a moment after putting my face into the FOV. I was only visible every so briefly, so the snapshot didn't capture me. I'd probably have to lower the anti-dither or some other delay to have a fast snapshot.
But the point is I had FD working! With snap, w/ email. What went wrong afterwards is a puzzle.
Is it h/w? I do have a asticky IRC filter mechanism. But that seems unrelated...
Is it a configuration mistake by me? That's where my money is!
Sorry to hear that.
I had it working, ever so briefly. Below is an example of an email I received a moment after putting my face into the FOV. I was only visible every so briefly, so the snapshot didn't capture me. I'd probably have to lower the anti-dither or some other delay to have a fast snapshot.
But the point is I had FD working! With snap, w/ email. What went wrong afterwards is a puzzle.
Is it h/w? I do have a asticky IRC filter mechanism. But that seems unrelated...
Is it a configuration mistake by me? That's where my money is!
Last edited by a moderator:
@Fastb i noticed the 2 boxes don't stay where I set them either. The weird thing is on the nvr I have to set the boxes in line detection. I thought maybe having the camera record would do it because when you set the schedule through the camera itself a green bar like record is what shows.. Didn't work. So far face detection is a bust.
Sent from my SM-N920T using Tapatalk
Sent from my SM-N920T using Tapatalk
Last edited by a moderator:
To see face captures you have to open on a single camera then you can click face icon. I don't think it will give ivs hits because face detection isn't in ivs category. FD is its own option and disables ivs if you turn it on.@JFire,
Thanks, I hadn't noticed the person silhoutte before.
In my case, it's greyed out. Even watching playback with faces, and FD enabled.
At the bottom of the playback screen, below the colored bars, is a legend:
Do you see orange in your timeline for a FD event?
I was seeing orange with tripwire and/or intrusion zone events.
Sent from my SM-N920T using Tapatalk
Last edited by a moderator:
Fastb
Known around here
@JFire,
Thanks for the "single view" tip. That solved my greyed out "face icon".
Then I noticed a button just to the right of the face icon button. It lists the events with a time stamp.
Most were events were M for Motion.
I noticed one F. Eureka!
Clicking on that event in the list brought me to the exact playback point for that F event.
I had a FD of 1 second! On my shrubs! In the dappled light, the camera mistakenly thought it saw a face.
It reminded me of something I read. I got the manual.
All 3 factors listed are very different from my lab (where FD worked) and outside (where I have problems)
I think my installation/environment may be the problem.
I think I meet this requirement: "the width and height of human face must be bigger than 100x100 pixel".
FD probably uses the full, high res image in the camera - so pixel count is unrelated to the encoding I configured. (reasonable?)
(I use 1928x1080 for main stream, 704x480 for sub stream). I probably meet the 100x100 pixel count. But I don't have a way to check that.
Does my head look grainy? lol!
I'm gonna check angle. Must be "0 to 30 degrees from horizontal". I may exceed that.
I'm on the hunt!
Thanks for the "single view" tip. That solved my greyed out "face icon".
Then I noticed a button just to the right of the face icon button. It lists the events with a time stamp.
Most were events were M for Motion.
I noticed one F. Eureka!
Clicking on that event in the list brought me to the exact playback point for that F event.
I had a FD of 1 second! On my shrubs! In the dappled light, the camera mistakenly thought it saw a face.
It reminded me of something I read. I got the manual.
All 3 factors listed are very different from my lab (where FD worked) and outside (where I have problems)
I think my installation/environment may be the problem.
I think I meet this requirement: "the width and height of human face must be bigger than 100x100 pixel".
FD probably uses the full, high res image in the camera - so pixel count is unrelated to the encoding I configured. (reasonable?)
(I use 1928x1080 for main stream, 704x480 for sub stream). I probably meet the 100x100 pixel count. But I don't have a way to check that.
Does my head look grainy? lol!I'm gonna check angle. Must be "0 to 30 degrees from horizontal". I may exceed that.
I'm on the hunt!
Last edited by a moderator:
Fastb
Known around here
@JFire,
Mine might be too high, and the downward angle might exceed 30 degrees.
Your might be too high! The manual said "0 to 30 degrees". Which caught my attention. Why not say + 15 degrees?
So the cam is looking at the horizon, or downwards by as much as 30 degrees.
You might try to raise your camera. Or knock the boys down!
Mine might be too high, and the downward angle might exceed 30 degrees.
Your might be too high! The manual said "0 to 30 degrees". Which caught my attention. Why not say + 15 degrees?
So the cam is looking at the horizon, or downwards by as much as 30 degrees.
You might try to raise your camera. Or knock the boys down!
Last edited by a moderator:
Last edited by a moderator:
Fastb
Known around here
@JFire,
I mispoke. Yours might be too low (not too high).
ie: camera is looking upwards at faces, not downwards.
Yours might be +15 degrees, not meeting the 0 to 30 spec. +15, greater than 0.
(I assume the 30 is negative, ie: below the horizon, ie: 0 to minus 30)
I'm re-positioning the cam in a moment. News soon.
I mispoke. Yours might be too low (not too high).
ie: camera is looking upwards at faces, not downwards.
Yours might be +15 degrees, not meeting the 0 to 30 spec. +15, greater than 0.
(I assume the 30 is negative, ie: below the horizon, ie: 0 to minus 30)
I'm re-positioning the cam in a moment. News soon.
Last edited by a moderator:
Fastb
Known around here
@JFire,
I updated f/w in the cam. It was 6/15, and new f/w is 11/15.
No difference.
Been working on angles and illumination. Repositioned the cam, I'm within the 0 to 30 degree window. Bright light and shadows are reduced at this time of day. Video has clear imagery of my face, suitable for FD.
I am getting more FD events (good), but very unpredictable (bad). If I get my mug into the cam, an FD event occurs.
I was within the 0 to 30 degrees window. I walked all the way from my driveway (small face) until I consumed the entire image (big face) Finally it triggered!
Using the Dahua manual, I think I ruled out their 3 guidelines;
#1) installation angle adjustment
#2) Face size.
#3) environmental illumination.
Conclusion:
FD works. I get texts, snaps, etc.
FD doesn't work. It's way too insensitive. I have to stare at the camera from 15' away, and when I'm within 1', finally, it decides "I see a face!"
I updated f/w in the cam. It was 6/15, and new f/w is 11/15.
No difference.
Been working on angles and illumination. Repositioned the cam, I'm within the 0 to 30 degree window. Bright light and shadows are reduced at this time of day. Video has clear imagery of my face, suitable for FD.
I am getting more FD events (good), but very unpredictable (bad). If I get my mug into the cam, an FD event occurs.
I was within the 0 to 30 degrees window. I walked all the way from my driveway (small face) until I consumed the entire image (big face) Finally it triggered!
Using the Dahua manual, I think I ruled out their 3 guidelines;
#1) installation angle adjustment
#2) Face size.
#3) environmental illumination.
Conclusion:
FD works. I get texts, snaps, etc.
FD doesn't work. It's way too insensitive. I have to stare at the camera from 15' away, and when I'm within 1', finally, it decides "I see a face!"
Last edited by a moderator:
Fastb
Known around here
@JFire, @nayr,
It's nearing the end of the today's T-Shoot and investigation into Face Detect (FD). Based on my experience so far w/ FD and based on JFire's struggles with same, here's my assessment:
- Dahua's Camera FD is not working well enough. Flaky.
- I can't evaluate FD to IVS to decide which is best for Job Site Security for my 1st customer. (IVS = tripwire, zone intrusion)
--- The problem may be with me (config settings, ignorant/newbie user) (I'll give Dahua the benefit of the doubt)
- Dahua manuals and guidance to DIY people trying to do it themselves do NOT provide adequate guidance
--- Using Dahua products, without an OEM, shouldn't mean we must struggle without scant info and manuals
Face Detect right now is unusable for "Job Site Security" for my first customer.
- I will drop back to tripwire or zone intrusion since Face Detect is so lame and dysfunctional.
Maybe things will improve in the future. Fingers crossed.
If anyone has advice, I'm all ears.
It's nearing the end of the today's T-Shoot and investigation into Face Detect (FD). Based on my experience so far w/ FD and based on JFire's struggles with same, here's my assessment:
- Dahua's Camera FD is not working well enough. Flaky.
- I can't evaluate FD to IVS to decide which is best for Job Site Security for my 1st customer. (IVS = tripwire, zone intrusion)
--- The problem may be with me (config settings, ignorant/newbie user) (I'll give Dahua the benefit of the doubt)
- Dahua manuals and guidance to DIY people trying to do it themselves do NOT provide adequate guidance
--- Using Dahua products, without an OEM, shouldn't mean we must struggle without scant info and manuals
Face Detect right now is unusable for "Job Site Security" for my first customer.
- I will drop back to tripwire or zone intrusion since Face Detect is so lame and dysfunctional.
Maybe things will improve in the future. Fingers crossed.
If anyone has advice, I'm all ears.
Last edited by a moderator:
Fastb
Known around here
@JFire,
With the camera back inside in my shop, FD is working better.
Still unreliable.
- At 3am last night, the camera saw a face in picture on the wall (a b&w pic of a city scape)
- It sometimes does a FD when I enter the shop. Other times not.
- I re-oriented the camera. It was mounted for cable considerations. I had to flip the image 180 degrees, and enable mirroring. No longer. FD performance seems improved.
- I played with the size of the target boxes. The default doesn't work as well as setting the boxes to the max size of a face, and the min size of a face. The performance difference isn't striking, however.
- FD takes time. Walking slowly helps. Looking into the camera helps.
- Angle seems to have some effect. It seems the camera is better at FD when looking downwards slightly. Like the above points, the differences isn't striking.
FD inside works better than outside. Distances are less, making the faces bigger. And the background is less complicated, walls as compared to nature.
Have you made progress? Or thrown in the towel? @nayr, same question to you.
With the camera back inside in my shop, FD is working better.
Still unreliable.
- At 3am last night, the camera saw a face in picture on the wall (a b&w pic of a city scape)
- It sometimes does a FD when I enter the shop. Other times not.
- I re-oriented the camera. It was mounted for cable considerations. I had to flip the image 180 degrees, and enable mirroring. No longer. FD performance seems improved.
- I played with the size of the target boxes. The default doesn't work as well as setting the boxes to the max size of a face, and the min size of a face. The performance difference isn't striking, however.
- FD takes time. Walking slowly helps. Looking into the camera helps.
- Angle seems to have some effect. It seems the camera is better at FD when looking downwards slightly. Like the above points, the differences isn't striking.
FD inside works better than outside. Distances are less, making the faces bigger. And the background is less complicated, walls as compared to nature.
Have you made progress? Or thrown in the towel? @nayr, same question to you.
Last edited by a moderator: