Not sure if this is what you are looking for, but I presented a live demo of this at CVPR a couple of years ago.  Unfortunately, the only paper that we have had time to publish on this is the ISVC paper linked below; which, doesn't tell you how we use it for tracking.  The problem is that we're not really focusing on the tracking aspects yet; so, we used a semephor lock on our data buffer.  Needless to say, a semephor lock should never be used with streaming data, as can be seen from the linked videos.

https://www.youtube.com/watch?v=PiBdE1VdFG4

https://www.youtube.com/watch?v=9ikWSi4ZiG8

http://www.springerprofessional.de/008---voronoi-diagrams-of-line-segments-in-3d-with-application-to-automatic-rigging/5478220.html

Hi Dear,

     I think the attached material will address your issue. If still you have some doubts feel free to ask.

Ramin, is your question field specific?  I ask because If it is field specific, such as the material covered in Asutosh's response, then it makes a lot more sense to me.  If it is not, then I am not really sure what you mean by gating?  I know that most tracking methods that I am familiar with require some form of training, and I had assumed that you were referring to something along these lines.  If you are not being specific to the medical industry then there are a ton tracking methods out there.  Virtually all of the methods that I have worked with rely on frame to frame feature extraction, and not the type of gating that Asutosh is talking about.  I mentioned the one earlier because it does not require training; which means that it could be more ideal or less ideal, depending on what you are trying to do.

That said, probably the most commonly used tracking method that I am aware of, with respect to sheer volume, is that used by Kinect.  But they are definitely not the only one out there.  They are the easiest to get information about though; you just have to like reading patents.

Hello. Thanks for your answers. 

Consider for example the great Joint Probabilistic Data Association (JPDA) filter.  In this method all of the measurements obtained from sensors (radar, camera, ...) aren't used for estimation and instead by using a gating mechanism, some of the measurements are selected because there may be clutters. 

And my question is about a method that uses another mechanism to select measurements or use all of them.  

It sounds to me like this might be a "camp" issue.  That is, it sounds like the research "camp" that you are coming from is used to starting with a large variety of sensors where a subset of one or more sensors are used as a "gate" to either determine the validity of the data from the other sensors or to turn on/off the other sensors during certain scenarios, or some other similar gating functionality.  The research "camp" that I am familiar with, and the one that it sounds like you are looking for, is the one where the researcher is used to starting with a single sensor device, such as a depth + RGB camera, streaming black & white video, streaming color video, or etc., and then trying to obtain as much information from that single device before introducing more devices.  Once they feel comfortable with their ability to track with a single camera/device they then add more cameras/devices that each independently compute their perspective of the scene, and then a centralized device, frequently a control device, collect all of these perspectives and merges them into a more complete picture of the scene.  This merge step is frequently done using a simple machine learning process to overcomplicate a majority select-style operation on the input data.

Another difference between these two camps is that the "single-device first camp" frequently try's to do things like merge two disjoint scenes.

Does this make sense?  If so, I can try to add some links from my so-called "research camp" later.

I don't think (association) gates are really required in PDA. They are mainly used as a computational saving and/or to ensure that approximations/assumptions/simplifications used in the probabilistic models (like non-uniform clutter or multiple targets etc) don't degrade performance. If the event probabilities in PDA are calculated properly, then distant measurement will be given a negligible weight, so they won't affect the state - in theory, it shouldn't matter if a gate is used here or not!

Have a look at the ref below (or others by them). It always uses N nearest neighbours. This works OK because they use a good clutter model and because the clutter is quite dense. The main motivation for this approach was so they could vectorize their code for parallel computation, i.e. so the same instructions are applied to each track.   

Colegrove, S.B.; Davey, S.J., "PDAF with multiple clutter regions and target models," Aerospace and Electronic Systems, IEEE Transactions on , vol.39, no.1, pp.110,124, Jan. 2003

Hello.

Dear Jeffrey what you've said does not sense to me. But I am graceful to your contribution and I learned from that. 

What Dear Hugh have said is related to my question. 

Hugh As i understand from your answer, all of the tracking methods can be used without gating mechanism. Is it true? And Is it practical?

Hello. The notion of a gate in single and multiple target tracking does not relate to multiple sensors per se. It refers to the statistical validation of measurements in order to correlate one or more measurements from a sensor with a given track in order to filter the measurement(s) and update the track. The process of gating is not essential from a theoretical point of view. In problems where there is a low or zero false alarm density, gating may not be needed. At low SNRs, and therefore in high false alarm densities (such as those used in radar and sonar systems), gating is essential for practical implementation of all tracking algorithms. Please have a look at the paper below that surveys different tracking methods. All of these would use gating in practice even if it is not mentioned in the algorithm description itself.

BTW, the JPDA / PDA filter is over-rated as a tracking technique. It continues tracking long after a nearest-neighbour Kalman filter would have lost the track by following false alarms. This is bad if you want track purity/low false track rate and good if you want track continuity.

Article Taxonomy of multiple target tracking methods

Random set based multiple target tracking methods, e.g. (C)PHD filters, do not use gating. The measurement elimination happens automatically through likelihood calculations.