How can I re-track an object if track(using optical flow) lost?

asked 2020-01-29 04:19:30 -0500

eLtronicsvilla gravatar image


What are the best approach/algorithm to re-track an object if track lost while tracking using dense optical flow? Here, both background and foreground is dynamic. ROI of Detected box of object is changing at different part of body of object just because of optical flow method. Note : Motion of the object is non-linear.

Can somebody suggest the good approach to handle this case. Thanks in advance. :)

edit retag flag offensive close merge delete


In neural networks this is usually done by using a lstm layer.

holger gravatar imageholger ( 2020-01-29 05:08:13 -0500 )edit

imho, you have to explain, how do you track an "object" at all, using optflow. ?

berak gravatar imageberak ( 2020-01-29 07:50:33 -0500 )edit

@holger Thanks for your response. In my case there is no neural network , only Image processing concept .

eLtronicsvilla gravatar imageeLtronicsvilla ( 2020-01-30 01:23:41 -0500 )edit

@berak It's detection per frame using LK-optical flow. So there in some cases I'm loosing detection like corner case , sudden stop case,turning case,etc. How can I deal this?

eLtronicsvilla gravatar imageeLtronicsvilla ( 2020-01-30 01:34:17 -0500 )edit

Use a neural network with lstm XDDDDD. OR.... I know that opencv has a tracker api - you can use this if the images are coming in frame by frame. I am not sure but i think i read somewhere that some classes of tracking api of opencv also uses some kind of neural network- but i am not sure about this - is also not important.

Try read some tutorials about opencv tracking api.

holger gravatar imageholger ( 2020-01-30 13:47:00 -0500 )edit