Camera to identify faces at 52' (in daylight)

[Lagerhead]

In any case, after resizing the image to 80x60 (actual size below), it worked out to net ~22 pixels between the center of each pupil:

This is an enlarged view of the same image in the application I used showing the measurement of pixels (NOTE THIS IS AN ENLARGED VIEW OF THE SAME PICTURE AS ABOVE):

But, but but... I think *THIS* is a truer (faithful to the pixels) enlarged view of the smaller picture (MUST ENLARGE FULL SIZE TO APPRECIATE THE DIFFERENCE)

 

[drvnbysound]

In any case, after resizing the image to 80x60 (actual size below), it worked out to net ~22 pixels between the center of each pupil:

This is an enlarged view of the same image in the application I used showing the measurement of pixels (NOTE THIS IS AN ENLARGED VIEW OF THE SAME PICTURE AS ABOVE):

But, but but... I think *THIS* is a truer (faithful to the pixels) enlarged view of the smaller picture (MUST ENLARGE FULL SIZE TO APPRECIATE THE DIFFERENCE)

My intention was to have the same image as I posted, but actually having background area as well with some surrounding imagery - not as you posted with that image enlarged to 640x480. If you were to be taking a full 640x480 image of someones face that close up, you would certainly have a couple hundred pixels between their eyes.

 

[drvnbysound]

IVB: I was trying to find the image you posted on AVS from your recent CCTV thread, showing the view from your front door camera - apparently you removed the picture, or moved it (likely with the new site update), because the link you had there no longer works.

I think that view is a good image of what I would personally want to see from the 52' distance. I think it would certainly give enough pixels on target (maybe on the high side) but it does have a decent FOV where you could possibly catch 3-5 or possibly more people at the same time, with enough pixels on target.

If you were zoomed in on a SPECIFIC area as much as shown above by Lagerhead, it would be practically impossible to get a view of a specific person at that specific spot.

Example (assuming you had the 2 camera setup with one showing the larger picture and one zoomed in enough as Lagerhead depicted above): You are trying to monitor kids at a bus stop location. With the larger Field of View (FOV) you could see whats going on - if they are indeed harassing people that day, but if you were zoomed in that much, there would be no way of telling if that kid is wearing the same clothes as the one doing the harassing, etc. Its just zoomed TOO much to give any identifiable evidence. The kid would have to stand perfectly on an X on the ground and be the right height to make sure you got his face. If 3 kids are wearing white shirts (seen by the SD camera) and you are zoomed in only enough to see their face and maybe even the collar of their shirt, how do you know which one actually did what?

You simply need a larger FOV from the hi-res camera - you must find the happy medium of getting enough pixels on target, while also being able to get a couple of kids in the FOV at the same time so you can actually identify who is doing what.

 

[Lagerhead]

If you were to be taking a full 640x480 image of someones face that close up, you would certainly have a couple hundred pixels between their eyes.

Right, but you missed my point. Yes the graphic file is 640x480 but the *content* in pixel terms is AS IF it were 80x60. When you enlarged the smaller image you (apparently) used a decent interpolation algorithm which improved on the small original and so lost the granularity, which I thought was essential to your point.

You simply need a larger FOV from the hi-res camera - you must find the happy medium of getting enough pixels on target, while also being able to get a couple of kids in the FOV at the same time so you can actually identify who is doing what.

That is the fundamental tradeoff.

 

[drvnbysound]

If you were to be taking a full 640x480 image of someones face that close up, you would certainly have a couple hundred pixels between their eyes.

Right, but you missed my point. Yes the graphic file is 640x480 but the *content* in pixel terms is AS IF it were 80x60. When you enlarged the smaller image you (apparently) used a decent interpolation algorithm which improved on the small original and so lost the granularity, which I thought was essential to your point.

You simply need a larger FOV from the hi-res camera - you must find the happy medium of getting enough pixels on target, while also being able to get a couple of kids in the FOV at the same time so you can actually identify who is doing what.

That is the fundamental tradeoff.

interpolation algorithm?? I simply used HP's Image Zone software, that is approx. 5-6 yrs old, to resize the image. They probably have something in there to reprocess a resized image... ???

In any case, if you can get a 80x60 image on the targets face, as I depicted - netting approx 20 pixels between their eyes, allowing the rest of the cameras view being large enough to give some spatial view of where/what you are seeing, *I THINK* there is enough data there to identify a person. I think I could certainly identify this person:

if they were put in a lineup of people at a particular bus stop.

 

[drvnbysound]

I just did another quick Google image search for people outdoors to find some other generic photos trying to encompass my thoughts..

I found this image (no idea who these people are, but thanks to them for posting a picture of themselves on the internet <_<):

Which is 640x480. With the view given there, there are approx 20 pixels between the eyes of the man on the far left (the only one I checked) and assuming the others are the same.

If you had that same view of the bus stop area, you would be able to view a pretty decent area, which should be able to show most all activity and give enough pixels to identify a suspect.

Me thinking again.... I think if you were to have an identical view of the bus stop in question, and showed that to a 13 yr old kid, along with the video showing him harassing people, and/or the other kids who were there, someone is going to talk. As far as I'm concerned, its undeniable.

 

[IVB]

IVB: I was trying to find the image you posted on AVS from your recent CCTV thread, showing the view from your front door camera - apparently you removed the picture, or moved it (likely with the new site update), because the link you had there no longer works.

Ooops, thx for the reminder, just restored the images.

Me thinking again.... I think if you were to have an identical view of the bus stop in question, and showed that to a 13 yr old kid, along with the video showing him harassing people, and/or the other kids who were there, someone is going to talk. As far as I'm concerned, its undeniable.

Dude, this is Oakland inner city youth we're talking about. No joke, they step up to the OPD, because after the Oakland Riders settlement (police brutality case a few years back), there needs to be directly witnessed criminal behavior. I've seen youth seriously bitch out OPD or loudly talk trast AT them, mocking them, and OPD can't do anything, and both sides know it.

Unless there is a clear video of the target youth committing these acts, and it's clear who they are, the situation isn't going to get better. Furthermore, there's concerns by some parties that the 13 year olds will be climbing up to the roof-mounted cameras, and destroying them within weeks. I'm going to have to armor (or disguise) this somehow.

 

[drvnbysound]

IVB: I was trying to find the image you posted on AVS from your recent CCTV thread, showing the view from your front door camera - apparently you removed the picture, or moved it (likely with the new site update), because the link you had there no longer works.

Ooops, thx for the reminder, just restored the images.

Me thinking again.... I think if you were to have an identical view of the bus stop in question, and showed that to a 13 yr old kid, along with the video showing him harassing people, and/or the other kids who were there, someone is going to talk. As far as I'm concerned, its undeniable.

Dude, this is Oakland inner city youth we're talking about. No joke, they step up to the OPD, because after the Oakland Riders settlement (police brutality case a few years back), there needs to be directly witnessed criminal behavior. I've seen youth seriously bitch out OPD or loudly talk trast AT them, mocking them, and OPD can't do anything, and both sides know it.

Unless there is a clear video of the target youth committing these acts, and it's clear who they are, the situation isn't going to get better. Furthermore, there's concerns by some parties that the 13 year olds will be climbing up to the roof-mounted cameras, and destroying them within weeks. I'm going to have to armor (or disguise) this somehow.

Those are indeed some serious conditions - sucks to have to deal with something like that. Im surprised that these kids are actually at the bus stop :-\

One case in particular that I have heard about was an outdoor view of a fairly well trafficked area from an over head camera. The camera was a SD analog camera, with standard lens. With the view that it had you could certainly not identify anyone specific from the camera, but had a good view of the overall area and what was going on. The camera caught a man in a particular colored shirt (red) doing some stuff he shouldnt have been. Shortly there after, an officer located a (the) man in similar clothing and questioned/arrested him. There was no specific proof that the man in the video was indeed the same man the officer found - however it was highly likely as it was in the same location within minutes. The video evidence stood up in a court case against the man. The only evidence that they had was the color of clothing the man was wearing on the video tape matched that of the man the officer arrested. With that alone, he admitted it was him and was sentenced accordingly.

Curious question.. if the kids that are doing this area as bad as you say. One of your earlier posts (#13) you said that you were hoping the principal of the school would be able to take action?? If they dont care about the OPD, are they actually going to respond to anything a principal can do?

 

[IVB]

Curious question.. if the kids that are doing this area as bad as you say. One of your earlier posts (#13) you said that you were hoping the principal of the school would be able to take action?? If they dont care about the OPD, are they actually going to respond to anything a principal can do?

Yes. The cameras can actually provide evidence of criminal or inappropriate public behavior, the kids are smart enough not to do that in front of OPD. They just act like pricks. With that proof, he has "expulsion" as the ultimate penalty.

The biggest thing that the kids would really hate is getting kicked out of the nice public school in the middle of yuppie-ville, and getting sent back to the hellhole that is their "neighborhood" school.

 

[drvnbysound]

Curious question.. if the kids that are doing this area as bad as you say. One of your earlier posts (#13) you said that you were hoping the principal of the school would be able to take action?? If they dont care about the OPD, are they actually going to respond to anything a principal can do?

Yes. The cameras can actually provide evidence of criminal or inappropriate public behavior, the kids are smart enough not to do that in front of OPD. They just act like pricks. With that proof, he has "expulsion" as the ultimate penalty.

The biggest thing that the kids would really hate is getting kicked out of the nice public school in the middle of yuppie-ville, and getting sent back to the hellhole that is their "neighborhood" school.

I follow all above. Good luck with the situation for sure.

I did want to add something... you stated that the kids are smart enough to not commit that sort of inappropriate behavior in front of the OPD. In the case I mention above, the officer who found / questioned the man didnt witness anything either. A bystander witnessed some of the incident from a pretty far distance (not allowing them to identify the person), went to the appropriate authorities and the officer was simply dispatched to that area for a man with that particular description. No officer saw any of the actions live in person, or live from the CCTV system. They simply went to the video files after the incident, and found they did indeed have video evidence of him in action. Matching the mans description from the bystander to that of the video footage and to the man who was found in the same area was enough evidence in that particular case.

I can honestly state that I am not sure how this occurred. If the video didnt have specific evidence pointing out it was indeed the man who was caught (ie good facial), I would think the man could simply state that it wasnt him, and was someone else who must have been wearing similar clothing. My only assumption is that the video didnt show anyone else in the area matching the description and the officer didnt see anyone else either - but that doesnt necessarily prove anything. Last I checked it was guilty beyond a doubt.. ?? :-\ Either way, not-so-great video is certainly better than none at all.

 

[amirm]

Why not go out and take a picture with any camera you have at multiple focal points and report back? That is how I tested what kind of lenses I needed for my CCTV cameras.

Most cameras also let you set the resolution. So you can choose say, 3 megapixels to match the Mobotix.

 

[IVB]

Why not go out and take a picture with any camera you have at multiple focal points and report back? That is how I tested what kind of lenses I needed for my CCTV cameras.

Most cameras also let you set the resolution. So you can choose say, 3 megapixels to match the Mobotix.

What a simple but brilliant suggestion, I should do exactly that to avoid all this guesswork. I can't make it to the location until next Sat, but i'll try that then.

Thanks.

 

[RichardK]

Axis 211M megapixel Network Camera $700ish

Axis 5502-221 5-50mm zoom lens, $240ish
http://www.axis.com/products/cam_5-50_megapixel/index.htm

Set up motion detection (built into the camera), FTP files to some machine with a large hard drive.

Axis lens calculators:
http://www.axis.com/techsup/cam_servers/lens_calculators/

 

[SnyperBob]

Hi, I was reading over this thread and was thinking that you could make a setup similar to how local gas stations video their pumps around here....

They have one or two cameras that take care of all the pumps. The camera just zooms in to where a car's license plate would be at pump one, then it zooms out and pans over to pump 2, then it zooms in @ pump 2 where the car's license plate would be, etc....

So when it zooms out you get a wide angle, and when it zooms in you get a close up of what you're pointing at. Since the video feed keeps rolling, you never miss out on anything and you don't have two camera feeds to compare or whatever.

You could just do a similar setup and PTZ the camera to all the common spots the kids stand at the corner. Even if the camera is zooming in on the wrong kid, when it zooms out you'll at least still have a bigger picture of the entire corner and still be able to see what's going on

 

[drvnbysound]

Hi, I was reading over this thread and was thinking that you could make a setup similar to how local gas stations video their pumps around here....

They have one or two cameras that take care of all the pumps. The camera just zooms in to where a car's license plate would be at pump one, then it zooms out and pans over to pump 2, then it zooms in @ pump 2 where the car's license plate would be, etc....

So when it zooms out you get a wide angle, and when it zooms in you get a close up of what you're pointing at. Since the video feed keeps rolling, you never miss out on anything and you don't have two camera feeds to compare or whatever.

You could just do a similar setup and PTZ the camera to all the common spots the kids stand at the corner. Even if the camera is zooming in on the wrong kid, when it zooms out you'll at least still have a bigger picture of the entire corner and still be able to see what's going on

Someone would have to either write custom code for this to precisely control the camera to automatically look at certain locations in a looped-like configuration, or special camera control software would have to be purchased separately. Certainly there are many people and/or companies who have written camera control software for the Pelco D protocol as well as some of the other well known protocols, but I doubt its cheap if purchased.

I think it would be better to have 2 cameras personally - what if an event happens while you have the camera zoomed in closely and you couldnt see the event occur?