[ANN] PawsomeTracker.jl

[yakir12]

I’m glad to share PawsomeTracker.jl, a simple, robust, and performant tracker used to track targets in video files. It uses a Difference of Gaussian (DoG) filter to detect the target in the frame.

It works with concurrency, videos that have a non-zero start-time, pixel aspect ratios (i.e. SAR, DAR, etc) other than one, and can process multiple consecutive videos (e.g. segmented video files). It returns a vector with the time-stamps per frame and a vector of Cartesian indices for the detection index per frame. It has a few useful settings with sensible defaults:

• start time
• stop time
• target width
• start location
• search window size
• brighter/darker than the background
• frames per second
• should a diagnostic video be saved

We’ve been using this tool extensively in our lab and it works very well, but feel free to suggest improvements, fixes, or anything really!

[alejandromerchan]

Looks like an awesome package. But I also just want to show appreciation to the fact that the Julia forum now has a video of dung beetles. Seeing the mix between Julia and entomology is exciting for me.

[yakir12]

Nicely spotted..! Indeed, it’s Kheper lamarcki.

[baggepinnen]

You’d be happy to learn that this has existed for 6 years already

Smoothing tracks with a Kalman filter

Of course! Go to GitHub - yakir12/CoffeeBeetles.jl: Code base for "A dung beetle that homes without the use of landmarks" (install but you don’t need to run main()), and then if you run: data = CoffeeBeetles.deserialize(CoffeeBeetles.datafile) df = CoffeeBeetles.getdf(data) that df is a data frame with tons of tracks. To access the raw coordinates of a track in the first row of that data frame, you can do this: t = df.track[1] t.rawcoords The smoothed ones (again via a smoothing spline) are…