Digital camera + IR LED + sensor = crop around objects based on distance?

[simplysped]

Hi guys,
This might be completely retarded but the idea popped into my head after seeing some of the xbox kinect stuff.

Imagine you have a point and shoot digital camera, but built into it very close to the lense of the camera was a powerful IR LED which would bounce of all of the objects in the photo. The sensor would pick up the approximate distance and would then be able to remove the background of an portrait for example.

Here's an illustration:

Is this possible? Has it been done? If not, why?

If it is possible, would it mean that you could make this effect with video without a costly greenscreen setup? There must be some reason why it can't be done.

 

[Tech_Junkie]

Umm, this is a laser forum. You might want to ask that question on an appropriate forum. Not saying no one will give an opinion, but questions like that should be addressed to people directly in the field.

 

[simplysped]

Umm, this is a laser forum. You might want to ask that question on an appropriate forum. Not saying no one will give an opinion, but questions like that should be addressed to people directly in the field.

Hmm alright. I guess I'll look for a digital camera board to bother

Thanks anyways

 

[Tech_Junkie]

I think you will have better luck in getting your answer.

Good idea, and good luck.

 

[bobhaha]

Maybe because it might be easier to drop a green screen behind the person and edit it out later. Also another thought came to mind... what if in the background there was reflective objects or objects on weird angles.. the pickup sensor will have a hard time figuring out where the heck the IR light went... while dynamically cutting out the background and measuring every pixel distance in the image. Lastly of course, because many things radiate IR light, our body, hot objects even air. This would all come into play in a movie scene.

Just a few ideas... just seems easier to drop a green screen in there...

 

[HIMNL9]

Also, remember that autofocus units works on contrasts ..... regardless if they use IR or visible light, an uniform background cannot be correctly picked as "distant object", if it don't have contrasting elements, and if the AF keep track of the object at 6 ft, canot correctly "cut away" the background only, where instead if it can correctly manage the background, cannot at the same time keep focused the front object .....

The exceptions are those high-cost professional camera objectives, that have so much resolution and depth of field that can show focused both the background and the front object, but in this case the software cannot detect a difference .....

Maybe you can use 2 different high-frequency-pulsed IR sources on the object, from different angles ..... the object is illuminated from both the sources, where instead the background have "border shadows" from one or the other source, caused from the object itself, so a software can detect the part illuminated from both the sources and keep it, discarding the rest ..... but i see it as complicated and high-cost system ......

Instead, a green or blue screen in the background can be easily filled with a simple chroma-key routine in post-production, giving them less costs and less problems .....

 

[AJ_Dual]

Edge detection is going to be a problem as well, because the sides of the head, the curve of the shoulders/torso etc. are going to be dark because only the surfaces of the subject that are relatively orthogonal to the IR lightsource will bounce back the most light. And as others are pointing out, things with an unusual albedo, like a suede jacket, or even a person's hair will give a false metric.

This isn't a problem for more crude applications like autofocus, but for actual object/edge detection it is.

So the tops of the person's shoulder, sides of their head etc. that should be inside the bounding edge might be too dark, or even darker than the background from which it should be separated.

Honestly, there's lots of great edge detection code already out there, and I think it's probably just easier to do it all through software/firmware with all the various pixel delta algorithms out there.

Photoshop, GIMP... etc. already does this pretty darn well and semi-intelligently. When you crop a person you just draw a very crude spline around them, and the software picks a good edge after that. Not like the old days in the 80's and 90's where you had to draw the line itself.

So considering that consumer-grade point and shoot cameras already have things like face and smile detection that's not based on any "trick" with a second light source, (it's pure digital pattern recognition, pixel-based formulas) what you want to do is probably already possible in real-time purely digitally. Its just that no one has thought to do it, or some product or software already does this, but it's specialized equipment for film and TV production or other vertical markets etc. to make a "virtual green-screen" camera function like that.

ETA: There's already borderline shovelware webcam software that does this, where the software just cuts the subject person out real time and puts them on top of a background.

http://www.audio4fun.com/webcam-morpher-screenshots.htm

http://www.shiningmorning.com/

It's not all that good, but it proves the concept, I'm sure if it's needed, TV and film industries or the vendors that supply them have the capability already to do it nearly perfectly.

 

[cd520]

How can anyone post that picture without saying "This is my boomstick!!" ?

 

[Toke]

Interesting idea, that a nearer object will reflect more of a hopefully unique camera supplied wavelength and can thereby be distinguished from background.

And yes, it is unlikely to be a new idea, have it's own practical problems, and that is why it is not used widely.
And T-J got a point, a photo forum could likely supply 117 elaborations on the problems with it.