Electronic Visual Assistant for Jetson Nano

Hello,

I have created this topic to explain my project EVA (Electronic Visual Assistant) for Jetson Nano.

EVA is a real-time system that mitigates atmospheric turbulence.

EVA provides real-time videos of Moon and planets :
jup_result

EVA provides HR images of Moon (pano) and planets :

Some views of the embedded system (8 hours of autonomy, touch screen, Jetson Nano 4GB ) :

Have a nice day !
Michel

3 Likes

Hello everybody,

The turbulence can be weak, normal or strong.
EVA (Electronic Visual Assistant) analyses in real time a sequence of continuous raw images to define, depending on the turbulence, a degradation model applicable to these raw images.
The nature of the turbulence may change from sequence to sequence.
The model which will be implemented in 2022, will be tested on the Moon, Mars, Jupiter and Saturn.
The first tests show that this method makes it possible to recover images even in very degraded conditions which is not the case, among amateurs, with current approaches.
On the other hand, the method also works for long distance turbulence and underwater turbulence.
When this method will be fully tested, at the end of 2022, an Linux on-board version will be offered to all those who experience the inconvenience of atmospheric and underwater turbulence: amateur astronomers, long-distance surveillance, wildlife photographers, underwater photographers, etc.
A website dedicated to GPU vision will be open in 2022 so that everyone can follow the operations in progress.

jup_captures

What will the fp16 computing power of Jetson GPUs be in 2023?

Michel

3 Likes

Your work is really great.

I will follow it for sure !

Alain

1 Like

Stacking of lunar and planetary images in real time is possible thanks to the calculation by GPU (graphics processor) carried out by the turbu software to which this site is dedicated.

turbu software is available under LinuxMint, it requires the installation of the CUDA Toolkit, ArrayFire, Guake and ZWO ASI SDK modules to run.

Visitor curious to test this new practice of astronomy (assisted visual) can download turbu software after having installed the necessary modules.

GPU-Vision site

1 Like

Hello,

i tried to download turbu.zip but it seems the downloded file is not the ggod one. Maybe i missed something.

Alain

Hello,

This version under Linux Mint 20.3 (Ubuntu 20.04) is made for a laptop with a graphics card at least equal to GTX 1050 Ti.
The version for Jetson will be released with JetPack 5.0 on Ubuntu 20.04.

Michel

yes but in the zip file, i get a file with no extension information. I thought it was a python file like turbu.py or an exe file like turbu.exe but i only get a file named turbu

Alain

No it’s a linux C/C++ executable, there is no extension.
You unzip it and run it from the command line in a terminal:
$ turbu
usage : turbu expo(ms) gain WB_R WB_B band_width high_speed k_quality
example: turbu 3 300 70 70 100 0 0.7

C/C++ code is optimized for Linux Mint 20.3 (Ubuntu 20.04) environment.

1 Like

ok, i understand.

Thx !

Electronic Visual Assistant and deep sky :

A degradation model is applied to left image. No dark and no flat but a mixture of synthetic noises to model the real noise.

2 Likes

Really impressive result !

Alain

Hello,
Electronic Visual Assistant process :

Have a nice day !
Michel

1 Like

Great work.

Did you use wavelets filtering ?

The final result is impressing.

I see you stacked 6000 frames. Maybe you could try to keep less frames (the 2500-3000 better ones).

Alain

turbu software automatically selects the best number of frames depending on the turbulence :

Yes, I use wavelets filtering, patch denoising and precise optical flow.

Latest version of turbu works at 180 fps (5400 raw frames stacked in 30 seconds).

2 Likes

Comparison with the image of an amateur astronomer :

The difference is that the image on the left (15-minute derotation) is obtained automatically and that the images of the assisted video used for the derotation are obtained live.
A single end-to-end processing software: turbu.
For the left image, telescope used is a C11 @ f/20 on EQ6-R Pro.

1 Like

The images (800x600) on the left are obtained live every 30 seconds (visual assisted) while the images on the right are obtained every 3.5 minutes (derotation of 7 images).
The visual aspect is practically identical.

jup_video

3 Likes

Another treatment of the video from August 11, 2021 with a more faithful rendering of details and colors.
jup_VA_1

With some nice details :

1 Like

New treatments :

Super resolution x2 :

More details :

I must say your Jupiter is terrific and your treatments are first class. Well done !

Alain