Meteor Detector
#1
I think a cluster of these would be perfect to comb through thousands of night sky photos every day to identify meteor strikes.

In proper terminology, my 'cluster' will not run MPI or other clustering software.  It will be a group of independent work nodes.  The basic workflow is:

Get three sequential images where the middle one has not yet been processed.
Process the photos (convert from Canon RAW format; adjust brightness, contrast, etc...; compare the images; look at middle image for meteor signs like streaks or other irregularities)
Report results back to the main MySQL server.

Other computers in this class (like Arduino, Raspberry Pi, etc...) just don't have the horsepower to do this well.  I have tried this on a single 64 bit computer running Windows 7 but this shows bottlenecks in processing, OS Context Switching and network/HD access.  I believe having several Pine64s crunch through the images will work much better.

If this work well and I have time and money for the additional hardware, I may use the same methodology to search for asteroids.  We shall see.   Wink
#2
(02-11-2016, 01:53 PM)BostonBay Wrote: I think a cluster of these would be perfect to comb through thousands of night sky photos every day to identify meteor strikes.

In proper terminology, my 'cluster' will not run MPI or other clustering software.  It will be a group of independent work nodes.  The basic workflow is:

Get three sequential images where the middle one has not yet been processed.
Process the photos (convert from Canon RAW format; adjust brightness, contrast, etc...; compare the images; look at middle image for meteor signs like streaks or other irregularities)
Report results back to the main MySQL server.

Other computers in this class (like Arduino, Raspberry Pi, etc...) just don't have the horsepower to do this well.  I have tried this on a single 64 bit computer running Windows 7 but this shows bottlenecks in processing, OS Context Switching and network/HD access.  I believe having several Pine64s crunch through the images will work much better.

If this work well and I have time and money for the additional hardware, I may use the same methodology to search for asteroids.  We shall see.   Wink

I love the idea of using Pines to detect meteors, brilliant! Smile
#3
I think the Pine64 would be a great unit for cluster/parallel computing, although I am by no means versed/an expert at all, just the bang for the buck, compared to the next step up of parallela(?), as you mentioned yourself. Sounds like a promising project. How many nodes/units would you run in a cluster?
#4
Initially I plan on four after my Proof of Concept is completed using one. I have my basic framework running on Debian. It uses Python to run ImageMagick using different parameters to detect meteor streaks. Typically I will have about 600 images each night in the initial phase (One 30 second exposure every minute for ten hours). Average processing time on my Debian machine is about 45 seconds per photo.

If that is successful, I plan to scale it up four times; as in four cameras; 16-20 PINE64s.

That's about a year in the future... First step... proof of concept. :-)

Oh...  for those curious...  Here is a sample input to my ImageMacick script.  (I didn't take this - just a sample I found on the net.)
[Image: m7.jpg]

And here is the output.  The meteor streak is pretty obvious.
[Image: _m7.png]
#5
Fascinating stuff! Definitely the most inventive prospective use for the Pine I've seen so far.
#6
What's the code look like for pattern recognition? I don't really deal with computer processing image scenarios, and after seeing how it's stripped down I can see what you're looking for as a human, but how do you make a computer recognize it?

Curiosity piqued!
#7
I started writing code in C# (what I normally write).  I knew it would need to be converted to C to get the needed performance from the Pine64 so started looking for tools that already exist.  Knew about ImageMagick but had no idea everything a person could do with it.  Basically, here is what I do to detect a meteor...

dcraw -T -W F123.CR2
convert F123.tiff -modulate 200% -colorspace gray -quiet -canny 0x1+5%+20% -hough-lines 1x1+60 F123.mvg

The first line uses dcraw to convert a Canon RAW image file to a TIFF.
The second line first brightens the image by 200%, then makes it monochrome, then uses an ImageMagick tool called the Canny Edge Detector, then the tool called Hough Lines is used to detect edges that are lines.

The Python scripts I am building, taking into account the image, will change brightness, Canny parameters, Lines paramaters and even contrast and noise removal to detect the meteors.   The final detection is simple - look in the final output .mvg file (which is just a text file with line coordinates) with lines that have similar slopes within a certain tolerance - and there you go.

Still trying to figure out how to filter out airplanes.   Big Grin


Forum Jump:


Users browsing this thread: 1 Guest(s)