Kalman filter and bounding box tracking

Hi all! I hope that someone can help me with that!

I have a video that shows a face. Firstly, I detect the region of the eye from a face. The detection gives me a boundind box with [x,y,height,widht] coordinates. So now, i need KALMAN filter to track this bounding box.

Is it possible with kalman or should i use another tracker? Also i need this tracker to be in real time.

Thanks in advance!

Answer 1

[Erik S.]

Hi,

Do you have computer vision toolbox?

There is vision.KalmanFilter method there you can use.

Answer 2

[Kostas Epp]

Yes of course i have the computer vision toolbox. Basically, i found how to give as input the bounding box as input but after the first loop the position [x,y] is lost. How can i have this point of the face tracked all the time???

Answer 3

[Dima Lisin]

The easiest way to use the Kalman filter is to track the centroid of the eye. If bbox is your bounding box, then you can compute the centroid as bbox(1,2) + bbox(3,4) / 2. Please see the Using Kalman Filter for Object Tracking example for how to track a single object. For tracking multiple objects using the Kalman filter, see the Motion-Based Multiple Object Tracking example.