My day job is web programming, and as I am fond of quoting to the business people "don't tell me how to fix the problem, tell me the problem and your wishlist and I will attempt to get as close as I can"
So, my camera module 'wishlist':
* removable or no IR filter
* M12 or CS lens mount
* 60 fps
* high resolution
* color not needed, only need B&W
My application wishlist:
a camera 'module' with a USB HID interface that can recognize blobs and fiducials and reports them to the USB host in a manner similar to a mouse or touchpad. The camera 'module' being capable of tracking object positions 60 times per second (60fps) and capable of recognizing 2-2.5cm fiducials. Fiducial recognition needs approximately 40px, so resolution needs to be 40px/2-2.5cm. Lens and focal distance will greatly impact the resolution, so having an interchangeable M12 or CS mount lens becomes very important.
I was thinking of using a RPi model A as the basis of this module with a camera module without an IR filter and a M12 or CS mount. From the previous discussions I understand could set the image processing to B&W, maybe adjust the contrast/sensitivity to eliminate IR 'noise' and then hand that off to the CPU for blob/fiducial detection.
I want to use a camera and some processing power to track infrared illuminated objects. I am building a interactive touch table, like the Microsoft Pixelsense tables (formerly known as Microsoft Surface) The second generation table is here: http://www.samsunglfd.com/product/featu ... elCd=SUR40
The second generation embeds IR sensors into the LCD module, the first generation used a separate projector and IR recognition system. The idea is that IR light is emitted from the unit and objects on or near the projection surface reflect the IR light back to a sensor. There are 2 main categories of IR recognition, 'blob' and 'fiducial'. Blob recognition is detecting the center and boundaries of the reflected IR. The blob and be from fingertips, a hand, or an object like a phone. Fiducials are comparable to detecting the center and orientation of a QR code.
The NUI group forums (http://www.nuigroup.com/forums
) are a gathering place of people building similar projects. The current solutions involve USB or Firewire webcams and processing software. The video processing CPU consumption is not trivial. For motion tracking the target is 60fps. For both methods or recognition the higher the resolution the better, however fiducial size is directly impacted by the resolution where higher resolutions supporting smaller fiducials.