Electronically Assisted Astronomy with a Jetson Nano

Alain, thank you so much for your contribution.
Good luck and best wishes for whatever you are going to do.
Hope to hear from you here some time again.

Best regards,


Will miss you. Totally understand not having time for side projects. Hopefully what you are doing is interesting. Thanks for you contributions!




i am very busy those months but i am not dead.

Still waiting for new hardware opportunities (camera, lens) and this will take some time.

But i try to improve my acquisition/treatment software when i can find some extra time.

Making sky survey can be a bit boring because most of time, you only see stars moving very slowly. Not really exciting. It is quite different from real time observation of a deep sky object.

So, i have to make a trigger to start (and stop) acquisiton when something special goes on and desappear (a satellite, a plane, something else).

It will be old fashion uninteresting routine (no AI) because i do not find suitable AI routine for that kind of thing (with so specific image).

Rendez vous dans quelques semaines.


1 Like

I made more tests with my selective amplification routine with old deep sky video and i fact, this routine could be useful. Maybe it will bring some interesting result.

Just have to wait some months to see if it is interesting routine, making some new real tests.


First test with satellite detection. Not really complicated but also not really simple.


1 Like

An other test :


Rather cool stuff! I have to wonder what it would be like if you had some sort of ballpark range finding on the objects (e.g., actually predicting the orbit). The average guy could never afford it, and something like a powerful radar would be out of question, but this had me thinking of stereo cameras. There are stereo binoculars designed for deep sky…also extremely expensive, and not such that they would give any kind of distance finding ability, but it makes me wonder what the AI for your tracking could do differently with low earth orbit and binocular vision (something you couldn’t even begin to experiment with unless you had lots of money though…the price is “astronomical”…yes, it is a bad pun).

Then again, perhaps AI would not have the same requirements to measure depth as what would be required for humans to find depth perception on low earth orbit objects via binocular…AI training has a way of eventually getting to the point of finding photographic information in situations where people can’t even tell there is a difference.

I have seen some motorized stereo huge binoculars/telescopes with a seat include in the system. It’s crazy equipment.

For sure, with a lot of money, we could make something really cool. 3D effect can come with a single binocular system and just one telescope. Brain try to rebuild 3D when he gets 2 pictures (1 for each eye). It’s strange.

For now, i don’t think AI can be useful. AI is just statistic routines which give one single result when you input a picture. You would have to train many models for deep sky. And noise will bring some issues.

Deep sky is really well known and we can also predict with very good accuracy artificial satellites coordinates. Mathematics are good enough for deep sky. The software i wrote to control the 2 axis motorized mount can calculate deep sky objects and artificial satellites coordinates. No problem with that.

With many time, big budget (2 good cameras, 2 very good lens like Sony 58mm F/D 0.95 8500€ each !, a good azimuthal motorized mount, a good head mounted stereo display), we could get something really terrific.

But this kind of system could only be afford by research or big company for promotional action or something like that.

The lens and the camera are the key points. Nikkor 58mm F/0.95 is a f*****g good lens but the price is terrific. With a very good sensor, it could be the ultimate system. Nvidia can give some support to guys like me but Sony does not care at all about guys like me.

I wait for IMX485 sensor camera, maybe by the end of the year. For the lens, i will have to stay with my actual lens. It is a big limitation but i have to be reasonable !


1 Like

I have a question i would like to ask to Nvidia CEO, Mr Jensen Huang : what is the prize for 8k views in a topic ?

It would be cool if someone here could forward my message to him.

So, here we go :

"Hello mister Huang,

first, i would like to say that your leather jacket is really cool.

Second, it would be quite interesting if Nvidia could invest some quite reasonable amount of money in a wide field deep sky survey system, with very good sensor, very good lens (wide open with quite some focal length), very good real time treatment of the acquisition using Jetson SBC and visualization of the real time movie in a VR head mounted display ; the camera is supposed to follow the head move).

Third, it would be useful if Nvidia could be involved in astronomy projects to help people like you and me discover the sky.

For sure, big date is cool and gaming cards are also cool. I am sure big data will help me knowing what to buy before i even thought about it and killing zombies is quite fun but astronomy is higher level for sure.

Please Mr Huang, think about using a small part of your money to make something really high level (i talk about the goal) !

Hoping for your answer soon !



I made a daylight test to detect things in the sky (mostly birds).

This time, i use detection as a trigger to start video recording.

The video of the test :


Same kind of test but this time, i add real time zoom on the target to identify it more easily.

I will modify my software to be able to follow 4 or 5 different targets (zooming).

I think this is really cool stuff.

Still waiting for Mr Huang response !


Hi Alain, really great results from the latest videos you posted - I have been thinking about something like that to scan the sky for birds (hawks). My wife has chickens and they need to watch out… :)

Your astrophotography work has inspired at lot of us at NVIDIA, it can be an exciting future with Jetson in space-related applications including telescopes, rockets, rovers, nanosats, ect.

1 Like

If one day, Nvidia wants to be involved in a really cool project, if i can help, i will (for free, i already have a job).

New test : multi target tracking (deep sky test) :

This will end the preliminary tests for deep sky.


1 Like

That is good tracking especially considering how small the original (unmanified) objects are. What kind of tracking algorithm are you using? Is it using background subtraction or optical flow, ect?

Hello Dustin,

as it is video, i work on the differences we can find in 2 frames. Just as my adaptative absorber denoise filter. Just have to apply appropriate threshold on the result of the 2 frames subtraction, detect the edges and isolate objects.


Cool addition for this version where it zooms in: Estimate what the actual zoomed picture would look like without motion blur. I suppose you have a reference to what direction and speed the camera is moving, perhaps AI could be used to remove the motion blur.

Oops: Looks like your last post is already onto that. Replied before reading your last post.

The camera is static. If the camera moves, i have false positive signals.


I see…this is why you mention 2 frame subtraction. I was thinking of an electronically controlled telescope mount earlier.


last night test with very large field view. Clouds can bring false positive detection but the result is not that bad.


1 Like

Daylight test :