VISION With Raspberry and AI tensorflow

[bernard]

Here the actual dev on ardumower vision and obstacle avoidance EVERYTHING IS IN DEV and need many ressource on Raspberry PI4 (but certainly possible to do better)

Process use actually AZURITBER and PIARDU to detect with artificial intelligence (Tensorflow) some obstacle present in the mowing area.

I stop to work with avoidance because in Bylane mowing,sometimes mowing finish to mow 1 hour on the 3 or 4 same lane.
I prefer to simply stop and reverse to be sure mower is never stuck.

Pi Camera is not able to detect distance or size of object, Instead it use score and size (in % of the full screen) to know if the mower need to stop or if Obstacle is too far and mower can continue.

Here the WIKI link:
AzuritBer Pi Firmware (English) - www.wiki.ardumower.de

and the video link:

 

[W1976]

Hi Bernard, is this the TAODI Ardumower?? I have seen your video on youtube, yesterday.

 

[lezabour]

really nice!!
adding some computer vision or IA would bring ardumower to another level

I tryed few months ago for another project to calculate distance / and size of an object, i use this:

Find distance from camera to object using Python and OpenCV

Learn how to compute the distance from a camera to an object or marker using OpenCV. Simple OpenCV + Python algorithm to find distance from camera to object

www.pyimagesearch.com

Measuring size of objects in an image with OpenCV - PyImageSearch

Today, I'll demonstrate how you can compute the size of objects in an image using OpenCV, Python, and computer vision + image processing techniques.

www.pyimagesearch.com

another way would be to put 2 camera, i think the PI4 can do it, but don't know about the load when using the softwares for the mower and tensorflow script...

also found this, don't know if it can help
https://github.com/rrichards7/Obstacle-Avoidance-CT (also found this maybe it can help)

 

[bernard]

really nice!!
adding some computer vision or IA would bring ardumower to another level

I tryed few months ago for another project to calculate distance / and size of an object, i use this:

Find distance from camera to object using Python and OpenCV

Learn how to compute the distance from a camera to an object or marker using OpenCV. Simple OpenCV + Python algorithm to find distance from camera to object

www.pyimagesearch.com

Measuring size of objects in an image with OpenCV - PyImageSearch

Today, I'll demonstrate how you can compute the size of objects in an image using OpenCV, Python, and computer vision + image processing techniques.

www.pyimagesearch.com

another way would be to put 2 camera, i think the PI4 can do it, but don't know about the load when using the softwares for the mower and tensorflow script...

also found this, don't know if it can help
https://github.com/rrichards7/Obstacle-Avoidance-CT (also found this maybe it can help)

Many thanks for your link.

Infortunatly In real life it's a little more complex.

In your 2 first link it's easy because you always have a reference size and use Open cv to compute distance with camera focal length.
In reality with AI detection we can detect a person standing or sitting or a child (so you don't have the reference size) and can't compute distance but maybe it can work on BALL or other object with fixed size

The last link is a simple OPEN CV mask algorithm that can certainly only work on snow and avoid everything that is not white (high contrast on image)

When i have time i try to work on code to detect faster (Using TF lite at 4 FPS or rewrite code to use C++ )

 

[lezabour]

yes you are right, always more complex in real life than in lab.
probably the use of ultrasonic /lidar/TOF sensor is better than computer vision to mesure obstacle distance
mix with AI/ vision to analyse video feed and get object (like your chair in video)

anyway very cool stuff