robotastic

This podcast is generated by community on OpenMHz.com curating interesting radio transmissions.

ha! yea.... The camera itself doesn't know when a building gets in the way. I have some general stop angles, but it is not perfect. There is some ML in there to get rid of photos like this... but I guess the model things that air conditioners look like airplanes.

In theory it should work with any camera the supports ONVIF, which is a standard for security cameras... actually, as long as the camera has an API you could probably get it to work. Since precision and image quality are important though, it could end up being more trouble that it is worth. You may have better luck trying to find a used one.

That is interesting! There is this cool piece of software that looked like it helped with a lot of this. It did visual object tracking with telescopes... but it is no longer available: http://www.optictracker.com/What_is_it.html

Alignment has been a big pain for me. I might actually have to look at how to use stars to correctly position everything.

I am not sure if it is actually useful... but if folks have ideas, I would love to hear them! I worked on it mostly to see if it could be done.

Totally - I think you could work back where the camera is based. I have the bearing / azimuth of the camera for each photo, so you could look up where each plane was at that timestamp and then project backwards.

I bet you had much better aircraft flying out of Hong Kong!!

It should work for helicopters. In DC, alot of them do not have ADS-B and the flight paths are a little different, so I haven't managed to capture one yet though.

That would be awesome! How fast can those mounts move? Would they be able to follow an aircraft? I have found that more zoom actually makes things a little tricker because you have to make sure it is leveled and aligned precisely. When my camera is fully zoomed, it has about 1 degree vertical field of view.

Thats amazing! If I lived anywhere near the water I would want to try and set something like this up.

I posted a bit more on what I am using on the About page: http://skybot.cam/about

I got all the camera gear on ebay, which was lucky.

I figure since it is all based on ADS-B data, there isn't anything too crazy about making it public. I am also still trying to figure out if it is actually useful for anything... besides getting unlimited plane photos!

I managed to get it for $500 on ebay... but I just checked and it looks like similar ones are around $1000 now, so I guess I got lucky. An Axis PTZ dome camera turned upside down works great too though and there are a lot of those available on Ebay.

I just upgraded the App Service plan I was using... It should be back now for a bit.

You could get away with a pretty cheap hosting bill, I think a $10 VPC would be more than enough. My current Azure forecast is around $65, but that is for a couple other things too. Half the reason I went big with the different features is that I just started at Microsoft and wanted to learn the different things... plus you get a $150 monthly credit.

All of the capture stuff just runs on a Raspberry Pi.

Definitely - Let me build a little better dataset. I need to augment a bit because it was false detecting on some of the contrails.

I built some scripts to also start building a model that can classify the different aircraft models. One of the interesting things with this, is that you get really accurate labels... probably better than what a human could do for images of this size.

https://github.com/IQTLabs/SkyScan/tree/main/ml-model

Good question - I just checked and my forecast for the month is $65... but I have a few other things running. One of the perks of working at Microsoft is that you get $150 monthly credit, so I didn't work too hard to keep the costs low. I could save a little by combining the API and App server.

Oh! that is a great idea. I am already tracking the camera's azimuth and elevation so I could probably composite everything to togther.

I have been using this Docker image on my Pi and it has been rock solid once you get it setup: https://github.com/mikenye/docker-piaware

oh yea! great point, I forgot about that. I am using PiAware to handle the decoding so I do have an Enterprise Account. I will have to check if I can make API calls for free.

Let me know if you ever go down this rabbit-hole. Happy to give you pointers! You can get some pretty decent cameras off ebay for <$500. Since you are close you do not need much zoom.

Doing a vision only approach would be really cool. A lot of the interesting planes do not have their ADS-B transponder on. The two systems could actually work well together. My system ends up generating a really nice, labeled dataset. I put some initial scripts / notebooks together for building ML models that can classify the different types of aircraft. https://github.com/IQTLabs/SkyScan/tree/main/ml-model/script...

Hi! I am the guy behind this, awesome to see it posted here. Happy to answer any questions.

To answer a question I saw pop up, I am using the FAA aircraft registration data to pull up the extra information about the planes. It lists who actually owns the plane, vs who is operating it. That is why you see so many banks. Places like Flightaware have much better data, but a pricy to look up.

Adafruit just released this as a product! https://www.adafruit.com/product/4800

This sounds awesome - If you have any documentation up on how to do this, I would love to point to it from the trunk-recorder wiki.

Hop on to https://gitter.im/trunk-recorder/Lobby if you are having trouble getting the https://github.com/robotastic/trunk-recorder software running. Trunk Recorder puts a wrapper around the OP25 software and lets you capture all of the audio from a radio system using an SDR.

It might be worth checking out https://www.assemblyai.com they let you build a custom audio model. One challenge with the audio from these radios is that it goes through some heavy compression. Traditional models will have a lot of challenges. Give a system that uses analog audio a try. The quality of the audio is a lot better.

Feel free to add something like this to some of the feeds from http://openmhz.com I can give you some pointers on how to pull down the streams of audio. They are already chopped into M4A files.

It would depend on what type of modulation they are using. If it is narrow band FM, it should work fine. There are probably lighter weight solutions though. If you wanted to use a RTL dongle, you would also need an upconverter, since the dongles can't capture frequency that low: http://www.nooelec.com/store/ham-it-up.html

It would probably work pretty good for that. If you were just trying to listen to one talkgroup at a time, the raspberry pi should have enough CPU. Let me know how it goes! I happy to help patch things up to get this working.

Anything with a TXCO is pretty good.

I like these: http://www.rtl-sdr.com/buy-rtl-sdr-dvb-t-dongles/ And these: https://www.amazon.com/NooElec-NESDR-SMArt-Enclosure-R820T2-...

There is also this Rust based P25 decoder that I have been wanting to give a try: https://github.com/kchmck/p25rx It is a ground up re-write that doesn't need Gnuradio, so it could be pretty efficient.