1
$\begingroup$

I read this paper and need your help to understand it.

Object Tracking with Improved Detector of Objects Similar to Target

In paper, particle at the frame $t$ is represented by: $$c^{(i)}_t(i=1,\ldots, N)$$ and after that say let $c^{(i)}_t$ used by the proposed method be defined as

$$(x^{(i)}_t, y^{(i)}_t, u^{(i)}_t, v^{(i)}_t, w^{(i)}_t, h^{(i)}_t)$$

where $(x^{(i)}_t, y^{(i)}_t)$ represents the center position of the target object, $(u^{(i)}_t, v^{(i)}_t)$ represents the velocity of it and $(w^{(i)}_t, h^{(i)}_t)$ represent the width and the height of the rectangle which involves the target object.


I can't understand if $c^{(i)}_t$ is particle and a particle is a pixel. How a particle in Particle Filter can have width and height?

How to set $x$,$y$ of the target for a particle?

$\endgroup$
3
  • $\begingroup$ Can you please edit the question for clarity and if possible, include a link to the paper you are referring to? $\endgroup$ – A_A Oct 2 '16 at 7:05
  • $\begingroup$ what paper you read !? please add reference link to your paper or add some content to your question! $\endgroup$ – Behzad Oct 2 '16 at 8:00
  • $\begingroup$ sciencedirect.com/science/article/pii/S1877050915023546 $\endgroup$ – ehsan rakhshani Oct 2 '16 at 8:31
1
$\begingroup$

I didn't read the paper but let me provide some intuition about object detection and tracking. When you try to track a target in a video, object detection algorithms might not be enough and you need to support the algorithm with target tracking methods in bayesian framework. It could be Kalman filter or particle filter. In particle filter or any Bayesian filter you first define your state which is composed of positions, velocities and dimensions in your case. You create random samples of states using importance distribution and use them to estimate the density of your state. You are kind of correct by saying that these samples are points of pixels in the frame. However, there exist more information in the samples. There are also velocities and dimensions inside those samples. The algorithm eliminate less possible samples and move forward with more possible ones. In this way, you obtain most probable pixels that object could exist. You also obtain most probable velocity and dimension for the object.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.