Moving camera object tracking using VLIB

Suhayil,

Background subtraction algorithms are intended for static cameras.

If you have a moving camera, please consider using Lucas Kanade algorithm for tracking.

Many of our automotive customers have successfully used VLIB's Lucas Kanade for tracking objects where the cameras in cars are continuously in motion.

Regards,

Senthil

hi Senthil,

thanks for your info, i'll try and get back to u.

Regards,

Suhayil.

Dear Senthil,

I tried VLIB's Lucas Kanade example you sent and it worked fine with 63% correctness. When I try the same with my own images, it crashes and the scenario is explained below.

- I have contigious frames from a moving camera video with a moving object in its vield of view, seperated as individual frames (640 x 480) . In the first

few frames (upto 15 frames), the object's dosen't move, and the size of the object is 64 pixels wide and 37 pixels high, and is centered at column 372 and

row 236.
- After the 15th frame, the object moves gradually and the camera pans to keep the object towards the center of its field of view as much as possible.

Therefore the object is moving and the background is also moving.
- In these frames upto the last frame, the object does not stay in the same position relative to each frame. The object's center moves towards column 465 row

311, and its size increases slightly ( 82 pixels wide and 39 pixels).

1.    How can I detect moving object with moving background, and give its information to LK "block" width and height?

2.    I stastically gave the possible mask area (166 pixels wide and 113 pixels high with center as column 432, row 275 ) as the LK "block" parameter. I 

also  resized all the buffers sizes to 2x115x170. Now the problem is when i process from the 2nd frame onwards, some of the values in NEWX[i] and NEWY[i],

become negative after calling VLIB_trackFeaturesLucasKanade_7x7 function.  This causes  the algorithm to crash when we try to use these negative values to

access the X[i] and Y[i] array.  My question is why these values should become negative.  Is this caused due to overflow.

I calculated that in the worst case scenario (with 4 levels of pyramid for object object mask size of 166 x 113), the  higest value stored in PoINTx[i] and

POINTY[i] is approxiately 2500, which is within the range of a signed short.  There should be no reasoon for the values to become negative.

Thanks & Regards,

Suhayil

hi,

can anyone clarify my doubt posted above or give a brief about how to detect and track moving objects in moving camera using VLIB's Lucas Kanade, Kalman Filter, and/or other APIs.

Thanks in advance,

Suhayil.

I successfully implemented VLIB's Background Subtraction for static camera and its working fine.

But I need to detect moving objects when the camera is also moving (i.e., the camera is mounted on a Pan-Tilt Unit).

How can I start detecting moving objects using VLIB's APIs? I got some idea from Lucas Kanade example for tracking the detected objects, but before that, how could I detect the moving objects when the camera is moving (where the background is also moving) So that I could give the info to Lucas Kanade or Kalman to keep tracking on it.

Can anyone give me a brief on how to use VLIB's VLIB_trackFeaturesLucasKanade_7x7 & VLIB_kalmanFilter_2x4_Predict with VLIB_kalmanFilter_2x4_Correct for moving camera object detection and then tracking the block.

I've been waiting for a long time just to get a brief to start with. Expecting someone to help me.

Thanks in advance,

Suhayil.

hi Senthil

could you send me a demo of  Lucas Kanade tracking using VLIB? 

my email address is : popfan229 @ gmail.com

Thank you very much!