Identify number of people in vehicle 2

[eddiebackman]

Hello!
We are two students that are going to start with our bachelor-degree thesis where the goal is to find and try the best possible method to identify the number of people in a vehicle in a roadside custom system which is mounted over the road. This is our first look at the project and we are trying to decide which method we are going to use to solve the problem.
We have come to the conclusion that a regular IR/heat camera won't be able to see through a window and that we probably will need to use some kind of “Near IR camera” to see through a windshield.
We are not yet that familiar about this technique. We gladly accept feedback or any tips about the project or a different approach on the problem.
We have decided that we are only focusing on the front seat ,to determine which soulution that will be potentially best.
Do you think that it's a possible solution?

 

[IRstuff]

It can be; window glass is transmissive up to around 2.5 um, so SWIR is possibly doable, but the sensors are expensive. SWIR has some unique properties w.r.t. human skin and most synthetic materials. Imaging sensors are 2D by nature, so they have limited abilities. One could consider high range resolution laser radars which could potentially get returns from within the car and some (non-trivial) algorithmic work could detect the difference between an empty seat vs. an occupied one.

Your application sounds like it's for HOV lane verification of occupancy, so the biggest challenge is cost.

TTFN
I can do absolutely anything. I'm an expert!
homework forum: //

http://www.engineering.com/AskForum/aff/32.aspx

faq731-376 forum1529

 

[itsmoked]

I think you would be better off using regular video with maybe a bit of filtering. The trick is to image the vehicle, assume there is a driver and look for close similarity with regard to the passenger side.

Keith Cress
kcress -

http://www.flaminsystems.com

 

[IRstuff]

The difficulties with imaging are myriad:

> Lighting is a big issue, night time imaging will require illumination
> Countermeasures; illumination can be somewhat defeated with reflectors
> Automatic detection is hard with relative simple targets like tanks; fuzzy-wuzzy, random, people under poor lighting is another thing altogether. Algorithms have gotten better, so face detection works, sometimes; other times, blocks of wood get detected as faces.

TTFN
I can do absolutely anything. I'm an expert!
homework forum: //

http://www.engineering.com/AskForum/aff/32.aspx

faq731-376 forum1529

 

[eddiebackman]

Thanks alot for the input , very interesting!

We got an answer from a store, they said that a possible solution was to use a cheaper and higher resolution camera : silicon-based CMOS.
They also said that you could use a "VIS block filter" so that only the "NIR"-part of the sensors sensitivity is used.
Then that we might needed a NIR lightsource to be able to use the camera in the dark.

what do you think about this approach? It is a huge difference in price compared to the regular NIR cameras.

 

[IRstuff]

The first 2 items on my list from the last post, as well as the 3rd.

What you describe is what almost all existing security cameras can do now. see

http://foscam.us/products/foscam-fi9821pv2b-megapixel-wireless-ip-camera.html

The only major difference is the range the illuminator will work over. You'd need something more like 50 ft or more in real-life.

Are you doing a proof of principle or an actual working demo? If the former, then you can look for cheaper cameras

http://www.ebay.com/itm/FREE-SHIP-T...667909?hash=item2ededf3dc5:g:qMwAAOSw6LBVA~zu

TTFN
I can do absolutely anything. I'm an expert!
homework forum: //

http://www.engineering.com/AskForum/aff/32.aspx

faq731-376 forum1529

 

[eddiebackman]

We are going to do a working demo in their roadside system. We haven't got any details about the budget, but i guess that it will be a decent ammount of money.


This is the hardware that we got recommended:

JAI GO-5000M-PGE
Monochrome area scan camera, Power over Gigabit Ethernet of the GO series
* 1"-CMOS-sensor with global shutter
* 2560 (H) x 2048 (V) pixel with 5.0 µm pixel size
* Frame rate 22 frames/s (at full resolution)
* Pixel-Clock 48 MHz
* S/N-Ratio >55 dB (0 dB gain)
* Digital output 8, 10 or 12 bits
* Power consumption via
6-Pin connector (12 - 24 VDC ±10%, 2.88W @ 12V) or

MIDOPT FIL LP715/30,5
IR Longpassfilter
* Cut-Off 715 nm
* filter thread M30.5 x 0.5

* incl. anti-reflective coating

A lens and a NIR led lightsource for the night.

 

[IRstuff]

Note that the camera cited is described as progressive scan, i.e., the image is not a snapshot, and there will be a lag between the first image row and each subsequent row. This would not be my first choice for such an application.

What is the distance between the camera and the car?

TTFN
I can do absolutely anything. I'm an expert!
homework forum: //

http://www.engineering.com/AskForum/aff/32.aspx

faq731-376 forum1529

 

[eddiebackman]

We are not 100% sure, but around 13 meters +- 5.
The camera will be mounted at approx 7,5m above the ground, that is also just an approximation by me.
We are in the start of the project and don't have that much information yet.

 

[IRstuff]

OK, at that distance, the illuminator will require optics, ala a searchlight, to just slightly overfill the field of view of the camera.

You elided over my comment about progressive scan. For a moving vehicle progressive scan means that different parts of the image will are not simultaneously acquired, result in possible distortion.

I didn't try to do the analysis, but you might want to consider whether the view of the occupants might get obscured by sun visors at the look angle dictated by your numbers.

TTFN
I can do absolutely anything. I'm an expert!
homework forum: //

http://www.engineering.com/AskForum/aff/32.aspx

faq731-376 forum1529

 

[eddiebackman]

When you say searchlight, do you mean the "NIR lightsource"?
Do you still don't think that the camera we got recommended will work with that conlcusion? Or do you have any tip on what characteristics to look for that suits our kind of application in a better way?
Thanks for the answers, really appreciated!

 

[itsmoked]

By searchlight he means a bright focused beam, in this case it could be invisible.

That camera is likely not too good for your application because the car is going to move a considerable distance over the course of a single frame scan which will have the affect of distorting the image away from reality. This distortion will be different with every different speed.

There are other cameras that grab a whole frame then read it out, more like a camera with a shutter.

Keith Cress
kcress -

http://www.flaminsystems.com

 

[eddiebackman]

Thanks again for the input, we asked about the progressive scan and got told that
Progressive scan with that sensor is like a camera with a shutter as you described what we needed. That all the pixels is exposed at one point in time and then "read out".
He said that with the interlaced scan the pixels got exposed in different points in time. And then distort the image.