Camera to identify faces at 52' (in daylight)

[IVB]

My neighborhood watch just got a storeowner across the street from a problematic intersection to agree to put up a camera so we can figure out which kids from the nearby school are creating a ruckus. They've asked me to select, procure, & install equipment as they have a very limited budget, and can't afford to hire someone.

I'll take pics later today/tomorrow/etc, but it's going to be across the street, perhaps 15' high. I don't have the exact distance, but the street is 4 cars wide (a 2lane road, with parking on either side), plus the sidewalk on the camera-mounting side. No idea what that translates to, what's a lane, 10' wide, so maybe 40-50' away from target, 15' high.

Damn, I can't believe I had to pull up wikipedia to remember how to calculate the hypotenuse, but using 50' and 15' gives me a 52' distance to target.

Q1) Any suggestions on the type of outdoor camera we'd need to identify a face at 52' away? Daylight only, that's when 99% of the problems occur.
Q2) Is there such a thing as an internet accessible single/dual camera DVR (maybe dual camera later)?
Q3) The PC or DVR won't be local to the store, it'll be across the street at the target location. Hence I need to use either a wireless video transmitter or an IP wifi camera.

Here's the kicker - the annual budget for the neighborhood watch is only $1200. I'm sure a few hundred has already been spent. We might be able to scare up a very small # of donations, but perhaps not. An HD camera would obviously be best, but i'm not sure we can afford that plus all this stuff. Of course, it's better to exceed budget than have a crap solution that doesn't work when we need it, so guidance there is requested too.

I'm looking to see if someone would donate a PC in which case we just need a cheap NV3000 for the recording, hence freeing up budget room.

Thanks.

 

[drvnbysound]

My neighborhood watch just got a storeowner across the street from a problematic intersection to agree to put up a camera so we can figure out which kids from the nearby school are creating a ruckus. They've asked me to select, procure, & install equipment as they have a very limited budget, and can't afford to hire someone.

I'll take pics later today/tomorrow/etc, but it's going to be across the street, perhaps 15' high. I don't have the exact distance, but the street is 4 cars wide (a 2lane road, with parking on either side), plus the sidewalk on the camera-mounting side. No idea what that translates to, what's a lane, 10' wide, so maybe 40-50' away from target, 15' high.

Damn, I can't believe I had to pull up wikipedia to remember how to calculate the hypotenuse, but using 50' and 15' gives me a 52' distance to target.

Q1) Any suggestions on the type of outdoor camera we'd need to identify a face at 52' away? Daylight only, that's when 99% of the problems occur.
Q2) Is there such a thing as an internet accessible single/dual camera DVR (maybe dual camera later)?
Q3) The PC or DVR won't be local to the store, it'll be across the street at the target location. Hence I need to use either a wireless video transmitter or an IP wifi camera.

Here's the kicker - the annual budget for the neighborhood watch is only $1200. I'm sure a few hundred has already been spent. We might be able to scare up a very small # of donations, but perhaps not. An HD camera would obviously be best, but i'm not sure we can afford that plus all this stuff. Of course, it's better to exceed budget than have a crap solution that doesn't work when we need it, so guidance there is requested too.

I'm looking to see if someone would donate a PC in which case we just need a cheap NV3000 for the recording, hence freeing up budget room.

Thanks.

While I know you are not running facial recognition software and comparing these faces to any sort of database, but it will likely take approximately 15-20 pixels between the eyes of your target to identify.

While I havent done it, there is a way to work this into your calculations - 20 pixels on target of 1' x 1' size at 52' distance and figure out what resolution of camera will be needed.

 

[IVB]

Cool, thanks. At least that gives me some sense of resolution required to make this useful. I really don't want to go through all this work, then have total crap (eh, it was a teenager, wearing white, skinny, black/white/etc, medium hair, not sure if they're wearing glasses or whether it's male or female, and that's all I can tell you).

 

[Lagerhead]

While I know you are not running facial recognition software and comparing these faces to any sort of database, but it will likely take approximately 15-20 pixels between the eyes of your target to identify.

While I havent done it, there is a way to work this into your calculations - 20 pixels on target of 1' x 1' size at 52' distance and figure out what resolution of camera will be needed.

To figure this you need specific data on the camera -- CCD sensor size and pixel density and the focal length of the lens. But it doesn't look promising for your project.

To put an upper limit on it, we can calculate what would be required for a decent still camera to get 20 pixels in 1" at 50 feet. (One critical assumption here -- that the "20 pixels between the eyes" means 20 pixels in each direction, and in this case, horizontally.)

It turns out that a Nikon D70 (a decent amateur/enthusiast camera) would need a 100mm lens minimum. I compute this from a field of view of 11' at 50' with a 100mm lens, using 3000 horizontal pixels. (It's been a while so anybody please check and correct me here.)

I think this is a highly optimistic objective for most motion cameras; plus, consider that at that magnification level, and with 3000 horizontal pixels which exceeds any inexpensive video camera, you still only cover an area 11 feet wide. What about the rest of the scene?

Edit sp

 

[drvnbysound]

While I know you are not running facial recognition software and comparing these faces to any sort of database, but it will likely take approximately 15-20 pixels between the eyes of your target to identify.

While I havent done it, there is a way to work this into your calculations - 20 pixels on target of 1' x 1' size at 52' distance and figure out what resolution of camera will be needed.

To figure this you need specific data on the camera -- CCD sensor size and pixel density and the focal length of the lens. But it doesn't look promising for your project.

To put an upper limit on it, we can calculate what would be required for a decent still camera to get 20 pixels in 1" at 50 feet. (One critical assumption here -- that the "20 pixels between the eyes" means 20 pixels in each direction, and in this case, horizontally.)

It turns out that a Nikon D70 (a decent amateur/enthusiast camera) would need a 100mm lens minimum. I compute this from a field of view of 11' at 50' with a 100mm lens, using 3000 horizontal pixels. (It's been a while so anybody please check and correct me here.)

I think this is a highly optimistic objective for most motion cameras; plus, consider that at that magnification level, and with 3000 horizontal pixels which exceeds any inexpensive video camera, you still only cover an area 11 feet wide. What about the rest of the scene?

Edit sp

NOTE: By 20 pixels between the eyes, I did mean 20 pixels total, from center of eye to center of eye - 10 pixels from nose bridge to center of one eye. Not 20 on each side totaling 40.

 

[beelzerob]

Geez, just use the same cameras CSI does. Problem solved. :blink:

 

[Lagerhead]

NOTE: By 20 pixels between the eyes, I did mean 20 pixels total, from center of eye to center of eye - 10 pixels from nose bridge to center of one eye. Not 20 on each side totaling 40.

Got it. So then cut my estimate in half. You can use a 50mm-equivalent lens and gain a 22-foot field of view, or reduce the horizontal CCD requirement to 1500 pixels. Still seems a stretch either way.

 

[carry15+1]

I'm not sure how effective it would be at 52', but I bought a camera some time back that may do the trick. It is a Sanyo DSR-C100 and was designed as a bank camera. It takes still images every few seconds at a resolution of 1.5 megapixels. It stores these on an internal 10 GB HD. IIRC it can record about 90 minutes of images - so it would probably be best connected to a trip like a motion detector. This camera comes with a slightly wide angle lens which may not work out well at 50+ feet. Just food for thought.

 

[drvnbysound]

Just did some quick photo editing. I pulled up a generic picture of someone's face on google. This was the original picture:

The original image was 210x158 pixels

I imported this image into a editing program I have, to resize (make it smaller) the image to a scale that gave the approx. specs I spoke of earlier so I could actually see what kind of images I was actually talking about - the information I had was directly from software vendors that I have spoke with before.

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):

 

[twilo123]

this is probably going to fall way short according to the calculations posted so far but mobotix has some good cameras to look at

http://www.mobotix.com/eng_US/Products

 

[IVB]

thanks for the replies. This is beginning to sound impossible, not only is this the only location option we have, vandalism is also a concern. I'll check into prices for the HiDef cameras, see how big&nice a camera I can get. I might be able to milk $800 or so for the actual camera, but man that vandalism Mobotix (for $3200) looks cool!

 

[Neurorad]

Not contributory at all, but can you describe the ruckus?

 

[IVB]

Not contributory at all, but can you describe the ruckus?

about why we want to target that intersection? A bunch of punk ass middle-school'ers from the disadvantaged parts of Oakland get bussed up here, and that's where the oakland bus stop is. They basically harass any people that walk by, and although they only occasionally get physical & actually touch people, on a daily basis they'll impede pedestrians movements and ask for $$, or other threatening gestures.

On a (weekly? every other week?) basis, there will be fights between kids, rarely is there an actual "crime" (ie mugging). Maybe 1x/month tops, even that's too frequent.

Hence the goal is to identify which kids are the bad apples so we can show pics to the principal, who can then take corrective action.

I've been considering something similar to what you're wanting to do IVB. Question, do you need hi-res video, or will hi-res stills suffice? I'm thinking of taking two cameras - 1, a cheapo usb webcam and 2, a middle-of-the-road digital camera and pointing them at the same area. The webcam will be monitored by either Motion on Linux or WebcamXP on XP and upon detecting motion within a pre-defined area will begin recording AND launch an external application. The external application will be a "trigger" program that will cause the digital camera to snap a photo or series of photos. This will give me both a low-res view of the scene and some high-resolution multi megapixel still of the area. As a test I took my daughters $150 Kodak digital camera and took a few pictures of the area in question, the resolution was great and I could easily read headlines on a newspaper that someone was holding.

I think the biggest challenge will be finding a (cheap) digital camera that will support a remote shutter trigger. I would preferably like something that could trigger through the USB conenction so that I could both trigger the shot(s) and download the images through the same cable.

I was going to start on this in a few days but the Ford guy just called me and my truck's repair bill is up to $3000 (diesel) so it looks like it will be on hold for a while...

Terry

REplying here as i think you meant to post in this thread.
As described above, i think video is important. Otherwise, any given picture won't really tell much of the story.

 

[personalt]

You could maybe check out some demos here
http://www.ipcamerademos.com/

I am not sure how high the roof is but looks to be a few stories.

 

[beelzerob]

REplying here as i think you meant to post in this thread.
As described above, i think video is important. Otherwise, any given picture won't really tell much of the story.

I almost wonder if you want a combination then....a fairly cheap camera that will give you the action via video, and the expensive still-shot-every-second kind to give you the detail.