Dear Abdullah, 

if the ball is sufficiently brighter than the background, I you could apply a thresholding algorithm to each image frame, set the limits (in pixels) of the minimum/maximum size of the ball and compute the centre of the highlighted area as the ball centre. You will also need to locate fiducial points throughout the pitch in order to reconstruct the 2D coordinates of the ball. You can find resources here:

http://it.mathworks.com/matlabcentral/fileexchange/29372-thresholding-an-image

and here:

http://it.mathworks.com/discovery/image-thresholding.html

From my experience in the early 2000's on this, tracking it isn't really the issue (lots of ways to do that). Rather the difficult part is recognising it. Modern GPU's are really adept at massive bitwise manipulations and comparisons so once you can recognise the ball in the scene, tracking its position is really the easy part. Are you using multiple cameras? Are camera(s) fixed position or free moving - or even motorised to follow? 

Cheers,Scot

Dear Abdullah,

Here is a simple solution (I like to call it simultaneous detection and tracking):

I am assuming the ball is colored (lets say Red). I am using opencv functions here to do image manipulations. The algorithm is simple and does tracking and detection simultaneously. This is a bit computationally intense, but the accuracy is guaranteed even with multiple balls. Also, I think there is a lot of room for modifying this.

Step 1: For every new image frame, repeat step 2 to step 5.

Step 2: copy the image to a new array and apply a median blur.

Step 3: Apply Hough circle Transform to the blurred image and overwrite the blurred image (from step 2) with the transform data to save active memory.

Step 4: for each circle detected in Step 3,  repeat Step 4(a) to Step 4(d)

Step 4(a): create a region of interest using the circle's diameter and center.

Step 4(b): create a mask using the region of interest from step 4(a)

Step 4(c): Apply the mask on the original (colored) image to get average color (for each channel) in the circular ROI.

Step 4(d): if the color is red. We found the red ball, exit the loop. The detection is complete!

Step 5: spit the result (i.e. The Hough circle enclosing the red ball).

look at the links for the programming how to's.

Cheers,

--Asheesh

http://stackoverflow.com/questions/10632195/define-image-roi-with-opencv-in-c

http://answers.opencv.org/question/10758/get-the-average-color-of-image-inside-the-circlehough-transform/

http://docs.opencv.org/3.0-beta/doc/py_tutorials/py_imgproc/py_houghcircles/py_houghcircles.html#hough-circles

Thanks everybody for the answers. I think my problem here is simpler. There are multiple cameras. Every camera is recording their own area in the field. There are intersecting areas in the field also (I mean two or more cameras are seeing the same  regions). The cameras are fixed, not moving. Problem here is finding the rough position of the ball, so that we can understand which camera should be recording (Assuming one camera is recording only, by a switching mechanism). 

Come to think of, to solve this recording problem, maybe only finding where the most movement is occuring (without finding the ball) would be enough. Any other ideas would be helpful.

You can put a tracking device in the ball and triangulate positions by positioning detectors in the field. Can you detect GPS signal in your indoor field ?. If so you can find the place where the ball is inside the field.

http://users.wpi.edu/~rsmolenski/Assignments/stem_wholething.pdf

You can track the ball with the TACTO software. You can find more datails in this paper.

https://www.researchgate.net/publication/47415367_Capturing_complex_human_behaviors_in_representative_sports_contexts_with_a_single_camera

Article Capturing complex human behaviors in representative sports c...