Tracking overlapping objects
Posted on : 13-05-2009 | By : Aleksey Kodubets | In : Demo, Demo video, Demo videos, YouTube
20
Description
This video demo illustrates color-histogram-based object tracker in action. CV system tracks people as moving blobs (“clouds” of moving pixels) identifies them and distinct one from another in case of occlusions. When two (or more) blobs are intersected, system merges them in one combined object and marks it by IDs of all those source-objects that currently included in the combination. When one of objects separates from the combination CV system recognize which one is out and re-arrange ID appropriately. This approach works pretty well in case of characteristic histograms.
Hi,
Will this work if all the subjects are dressed similarly, like in a soccer match?
Thanks,
– Pannag
Hi Pannag R Sanketi,
If it is important question for you and for your business case, we are ready to analyze your use case with more details and adapt our technologies for your needs. For soccer team, I could suggest a face tracker only with high resolution video which should work good. If you would like to have it, you can discuss with our marketing team (see “about” page for details).
Let me know if you have any more questions.
Aleksey
At the time of occlusion, how can you say certainly that blobs merged ? They might have gone out of the camera view especially if you are dealing with occlusions towards edge of view.
Hello Jithendra,
In the case of edge of view we can not say certainly that all merged blobs are still here or some blob have gone out of the camera region. We support this case as “guessing”. In common words each object has ability to be located in several probable positions
Hi may i know what kind of background and foreground subtraction you used? can you say that?
Hi Nikkey,
We tried Running Gaussian Average, Kernel Density Estimation, Mean Shift etc. This method on the video is own one and uses combination of mentioned methods + heuristics. if you have business use case, we could discuss with you and find best solution for your needs, see “about” page for details.
Thanks, ya ihave seen those algorithms in your website. I have to know, is it possible to do those things in real time. Wont it take much time? In my point of view “Running Gaussian Average” will take some time.
Nikkey,
Yes, it is real-time algorithms as you can see on videos. The algorithm is specially optimized for real-time performance.
Hi,
You have removed the moving shadows accurately. Have you done anything special works to remove the shadows?
Thanks
Kokul
Hi Kokul,
I clearly see shadows on the videos. Although, fps of the youtube video could be better. You can download overlapped_sonicfire.avi to see original video. If you have any particular question, you are welcome.
Aleksey
Hi Aleksey,
So isn’t shadow needed to be removed to track the moving objects? if the shadows are not removed, some times there are possibilities to system take the moving shadows as moving objects??
Kokul
Hello Kokul,
You are right. Shadow is needed to be removed.
We have implemented “color distortion” for background substruction to remove shadows
i would really like to have the source code for object tracking under occlusion in opencv + c/c++.
kindly reply me on my mail.
i would really like to have the source code for tracking overlapping objects in opencv + c/c++.
kindly reply me on my mail.
hi, am doing a similar project to detect and count people, can you help me in removing the shadows and defining each moving blob individually due to overlapping ??
kindly reply on my email .. thank you
have u find any solution for multiple people tracking & occlussion problem???
if u find plz send me demo code for this ..
thank u in advance
Does anybody have an idea about how to separate or detect overlapped objects? I am currently working on a project aiming at the detection crops in the late stage of growth. I did the first step, which is the segmentation (segment the plants from the background), but I got stack with detection stage because of overlapping between two crops or more (cauliflower). By the way, I am working on Opencv and visual C++. I will be grateful for your help.
have u find any solution for multiple people tracking & occlussion problem???
if u find plz send me demo code for this ..
thank u in advance
hi, i am doing multiple people tracking i am sucessdul in background substraction , find countours over moving object & tracking .but problem occur when two people are ovelapping please give me the solution or best algorithm for occlussion problem & send me some demo code related this i am working on c++
plzz rply fast if possible
thank u in advance ….
Hi,
I’m working on vehicle type recognition using videos of vehicles at a toll-gate. I mean to first detect each vehicle and crop it, after that I’m going to extract features and do some classifications and so on. I’ve tried to use Mixture Of Gaussian but it didn’t give proper results since there were cases in which vehicles can’t move enough before they are cropped and they are also occluded by vehicles at the back of them,do you have any idea about this? thank you.