yellow point -observation (detection)
blue point- predicted value (prediction phase)
pink point- estimated value (after measurement update)
initial state x = first detection location (blob center)
Pictures from 1 to 4 represent first frames.
1.As you can see in the result (pic 5) . It took more than (10 values) for both (predicated and estimated state) to get closer to the measurement values ( center of blob from detection)
2. The first prediction for (x,y) was (0,0) although Kalman filter has the initial state x .Is there any problem ?
Dear Aeymen Hassan,
I suggest you to see links, attached files and publications in topics.
GitHub - skhobahi/Kalman-Filter-Object-Tracking: 2D Object Tracking ...
https://github.com/skhobahi/Kalman-Filter-Object-Tracking
Using Kalman Filter for Object Tracking - MATLAB & Simulink Example
https://www.mathworks.com › ... › MATLAB Examples › Tracking
Best regards
Conference Paper Visual tracking of unknown moving object by adaptive binocul...
Conference Paper Real-Time Tracking of Moving Objects Using Particle Filters
Article Tracking a Moving Object by Visual Servoing
Article Real-Time Motion Tracking for Mobile Augmented/Virtual Reali...
Article Effect of depth information on multiple-object tracking in t...
Article Real-Time Motion Tracking for Indoor Moving Sphere Objects w...
Thanks, I found the problem in my test code (openCV c++)
I checked the source file in the opencv in:
opencv_files\openCV\modules\video\src\Kalman.cpp
It helps me for both solving my issue and understanding some points
The solution :-
change these equations
KF.statePre.at(0)=detectLoc.x;//intial state
KF.statePre.at(1)=detectLoc.y;//intial state
to
KF.statePost.at(0)=detectLoc.x;//intial state
KF.statePost.at(1)=detectLoc.y;//intial state
and now it behaves in the right way
In transition matrix , dt =1,the video is 30 frames/second. I noticed that if i change dt=1/30 both (predicted and estimated state value) become slow or (further away from observation point)
I thought setting dt equals to the time between frame sequences would improve the tracking result .
don't know Why this happen
- Badges
- Science topic
- Similar topics
- Correction