Dismiss Notice
Join Physics Forums Today!
The friendliest, high quality science and math community on the planet! Everyone who loves science is here!

Line Scan Camera for less than £30 ?

  1. Aug 5, 2015 #1
    Hi all,

    I'm looking to get started with a multitouch project using stereovision. Since I'm only interested in 2D for the moment I'm looking for a cheap line scan camera that I could use for the project. Ideally with an Arduino or that I can directly plug in the computer.

    The idea is to put 2 cameras in the top corners of the diplay that will be able to detect a finger touching the screen. Using the length of the finger image and a bit of trigonometry I can work out the position of the finger.

    More information on the project can be found here: https://www.physicsforums.com/threads/optical-touch-screen-code.825527/#post-5188484

    Does anyone know where I could find such a camera. Also what is the difference between a line scan camera and a line scan sensor ?

    Cheers !
  2. jcsd
  3. Aug 5, 2015 #2


    User Avatar
    Gold Member

    Why don't you use a "normal" camera that can see in two dimensions? A one dimensional camera will see a lot of things that it will regard as a finger. But you can learn a two dimensional camera, what a finger looks like: . So it must see a half circle with two parallel tangents added: That's a finger! This is not a finger:

    Due to the background in the picture, the one dimensional camera may not see an edge of the finger, but the two dimensional camera will.

    Say you want the camera to find a circle : O , but due to a missing edge it will see: C :
    The one dimensional camera, scanning horizontally, will not find the circle, but the two dimensional camera easy finds it by means of a Hough transformation. ( Look it up ).
    Last edited: Aug 5, 2015
  4. Aug 6, 2015 #3


    User Avatar
    Gold Member

  5. Aug 6, 2015 #4


    User Avatar
    Gold Member

  6. Aug 6, 2015 #5
    Thank you for your replies.

    My programming skills are very limited so I though that dealing with less inputs would be easier.

    My idea was that I would be able to get a row of values that I could then interpret: say 0000148410000000000000000 , each number is the value at one pixel along the line camera. I can then work out that the finger is located at the 14841 and centred on the 8. I'm aware that's a very simplistic view and that I will have to filter out the background noise but I thought it could work. Then, knowing the focal distance of my camera, the angle of my camera axis (45degrees from my x/y display coordinates), and the position of the finger on the image I can work out where it is using two cameras. (see here for more details: https://www.physicsforums.com/threads/optical-touch-screen-code.825527/#post-5188484).

    One thing I'm not sure about is will the camera be able to focus on fingers just in front of it (say 5 cm away) and at the other end of the display diagonal (say 85cm). or will I get blurry image at the extremes ? say 0011144488444111000000000 which would make the detection of more than one finger more inaccurate, as well as locating the actual centre of the touch.
  7. Aug 6, 2015 #6


    User Avatar
    Gold Member

    I'm worried about your span: 5cm . . 85cm: Yes, you will get some very blurred fingers, and say that the 5cm-finger takes up 14 pixels (11144488444111) then the 85cm-finger will take up 0.82 pixels ( or is it just noise? ).

    How do you know, that your pixel-patterns will be symmetric? Remember that the image is a result of a shadow/reflected light. Your attached link suggests no light-setting:

    No light → no image.

    I could suggest parallel light, created by some small light-bulb and a freshnell-lens. As this parallel light meets frosted glass, the shadow from a finger will always be seen sharp, and the distance between the camera and the frosted glass is constant.

    Sunshine is parallel due to the distance of the sun, thus a sharp shadow.
    Last edited: Aug 6, 2015
  8. Aug 6, 2015 #7
    I'm not sure what is needed to be able to focus over a span of 80cm. I'm worried that a mechanical lens will just not cope with the rate at which it has to adjust to gather data (30 FPS or around that).

    For the light issue, I thought that ambient light would be enough, just as for us when we turn the lights one. Another option I thought of would be to have a layer of glass above the display and shine IR LED through it. The camera field of view will be just above the glass but it won't detect anything since the IR will be dispersing upwards mostly, yet when a finger is put close to the glass its tip will be illuminated by the scattering IR and be more visible to the camera. A bit like looking at a screen from the side, you don't see the light coming out of it that much, but when an object is put in front you can see the light on it. Alternatively the light from the display itself might be enough.

    I wonder how they got around the focus problem in the paper though ?
  9. Aug 6, 2015 #8


    User Avatar
    Gold Member

    But emits your display/screen constant light?

    It's easy to test the setup: Just act a (stupid) camera at different circumstances ( different light/background, black & white color, and so on ):

    Can you always sort out a finger? Forget during the test your know-about that it is your finger you are looking at. :smile:
  10. Aug 6, 2015 #9


    User Avatar
    Gold Member

    And remember to look through a crack.
  11. Aug 6, 2015 #10


    User Avatar
    Gold Member

    Sorry: fresnel-lens.

Know someone interested in this topic? Share this thread via Reddit, Google+, Twitter, or Facebook