Detect an object, track that object while learning it. [closed]

asked 2015-04-23 17:12:35 -0500

215 gravatar image

updated 2015-04-23 18:46:58 -0500

I know that using the Haar cascades i am able to detect objects.
Using lukas kanade i am able to track an object. Lukas kanade is though prone to lose track of the object which means it has to be re initialized using haar cascades or other form of detecting, but it is not always capable of detecting objects (depending on how the object is posed).. does there exist a way where you using Lukas kanade can train/improve an existing a classifier such that when the object has to be redetected, it will be redetected using a improved classifier, such that it is capable of detecting the object in different poses?

Adding the tag Machine learning since it seems appropriate in this situation.

edit retag flag offensive reopen merge delete

Closed for the following reason question is not relevant or outdated by sturkmen
close date 2020-10-07 13:06:09.495587



It seems like OpenTLD seems to be able to do so, though am I not sure on how it works.. some form of clarification would be appreciated.

215 gravatar image215 ( 2015-04-23 17:55:57 -0500 )edit

It seems that OpenCV 3.0 has a contrib module with an implementation of the TLD algorithm. More information from the PhD thesis of Zdenek Kalal or his publications.

Eduardo gravatar imageEduardo ( 2015-04-24 03:22:05 -0500 )edit

I've tested the program it seems to work better , but not perfect... But about machine learning anything you could provide some info on... Unfortunately I Am not using 3.0.0 and don't think I will be able to upgrade it..

215 gravatar image215 ( 2015-04-24 04:35:00 -0500 )edit

Like said in your other topic (please try to combine questions together ...) TLD tracker in OpenCV is far from working decently. Better to use the original source code, right here. There is however a C++ implementation available also.

StevenPuttemans gravatar imageStevenPuttemans ( 2015-04-24 08:01:50 -0500 )edit

From TLD it seems to be the L part that i need to add. The T seems to be the same as i am using at the moment.. The code is just hard to understand..

215 gravatar image215 ( 2015-04-24 08:56:30 -0500 )edit

To understand the code, you will need to grab the original paper and start reading. None of this work is easy to understand, there is a reason why this is still a hot research topic :D

StevenPuttemans gravatar imageStevenPuttemans ( 2015-04-25 03:26:01 -0500 )edit