Trying to track an object (a fish) using Motion-based Multiple Object Tracking and am having troubles tuning the parameters to track just the fish and not ripples in the water.

2 views (last 30 days)

Graham_Neely on 11 May 2015

2
Link

Direct link to this question

https://www.mathworks.com/matlabcentral/answers/216278-trying-to-track-an-object-a-fish-using-motion-based-multiple-object-tracking-and-am-having-trouble

Commented: Dr.Rajesh T.M on 14 Apr 2018

I am using the code from the MATLAB site:

 http://www.mathworks.com/help/vision/examples/motion-based-multiple-object-tracking.html

to track a fish swimming in a raceway. I have videos filmed during daylight, and during night time (with infrared camera). I am hoping to track the fish's movement and eventually analyze the total distance traveled during the trial and hopefully also plot the fishes movement during select sections of the video segment. I am trying to tune the parameters from vision.ForegroundDetector to help track just the fish and not the ripples made in the water from the fish swimming. I am not very familiar with Gaussian Mixture Models and am therefore just blindly trying to tune the parameters (NumGaussian,NumTrainingFrames,MinimumBackgroundRatio) through trial and error. Does anyone have some ideas about what I can do either with the algorithm from the Object Tracking code from MATLAB, or with post processing of my recorded videos to make this tracking more effective? Attached is a set of the Object Tracking functions compiled together (ObjectTracking_DayControl). Also attached are screenshots (1 daytime, 1 nighttime) of what the videos look like (video file sizes are much too large to upload). The fish are small, slender objects in the picture, sometimes hard to see.

Thank you kindly, in advance!