Lego Brick Finder with OpenMV and Edge Impulse | Digi-Key Electronics
HTML-код
- Опубликовано: 20 сен 2020
- In this tutorial, Shawn shows you how OpenMV and Edge Impulse can be used to create Lego® brick finder.
Searching through a pile of Lego bricks to locate a particular piece can be a time-consuming task when you are constructing your Lego design (lovingly known as My Own Creation, or MOC). To alleviate that arduous task, we present you with the Lego brick finder!
In the video, we snap lots of still photos using the OpenMV camera module and crop out small sections of each image. We train a neural network using Edge Impulse to classify these sub-image chunks as either containing our target piece or not.
Then, we deploy the trained machine learning model to the OpenMV to locate our target Lego piece in each image it captures. Each time we snap a photo with the OpenMV, our program moves a sliding window across the whole image, computing the likelihood that each cropped section contains the target piece. If the likelihood is above our threshold, the part is highlighted on the LCD.
Please note that this is a proof-of-concept demo project with many limitations. Namely, it is very slow, taking around 10 seconds to identify parts in each photo captured, has a limited field of view, and works with only 1 target piece at a particular distance with particular lighting. Scaling this to work with all Lego bricks would require more time, more data, a large database, and a faster processor for the user.
Even though this is a fun demo showcasing machine learning on embedded systems, sub-image recognition and classification has many possible industrial applications. Such uses include self-driving cars, satellite image analysis, and X-ray image analysis.
Code for this project can be found here: github.com/ShawnHymel/openmv-...
Product Links:
OpenMV H7 Camera www.digikey.com/product-detai...
or OpenMV H7 Camera PLUS www.digikey.com/product-detai...
OpenMV LCD Shield www.digikey.com/product-detai...
Related Videos:
What is Edge AI?
• Intro to Edge AI: Mach...
TinyML: Getting Started with TensorFlow Lite for Microcontrollers
• TinyML: Getting Starte...
Getting Started with the OpenMV Cam: Machine Vision with MicroPython
• Getting Started with t...
Related Articles:
What is Edge AI? www.digikey.com/en/maker/proj...
TinyML: Getting Started with TensorFlow Lite for Microcontrollers www.digikey.com/en/maker/proj...
LEGO Brick Finder with OpenMV and Edge Impulse www.digikey.com/en/maker/proj... 
Наука
I'm 1 minute into this video, and so far this is some of the most relatable content I've ever encountered.
That is completely awesome! That’s a great idea. I hope this can somehow get community involvement. The possibility to put together an end to end video tutorial series by one or more persons on all the different aspects is remarkable. AI, programming, microcontrollers, photo/video concepts, IOS/Android app programming. Wow.
YES!!! A new learning video! I love the clarity and helpful information from Shawn. Thank you!!
Most welcome :)
When I saw the title of the video my first thought was "I already have two! My feet!". It's inevitable that when you have kids you will step on Lego bricks. Awesome project. Less painful than using my feet!
This is inspiring Shawn. Thanks!
Great job Shawn! Not only did I find this video entertaining I found it very educational. Hopefully a third-party or Lego will make an app for this really quick.
Why did your training data contain the "target" lego piece and noise (non-target) pieces in the same image (even though cropped)? Would it have been better to just use multiple images of the target piece and separately multiple images of noise pieces? I am new at this and trying to absorb as much as I can.
Which pin attached to push button pls help me
Can we use this data and models in Raspberry Pi ??
Actually because I don't have an OpenMV board,
I'm curious how I can use this tutorial for image processing in Raspberry Pi ??
Don't worry...I'm doing object detection for the Raspberry Pi next month :) I learned about single-shot detectors, which is a *much* better way of doing this project. In the meantime, I recommend checking out Edje Electronics' channel, as he's got some great object detection tutorials on the Pi: ruclips.net/channel/UCLuS8eZl3_nKKq85gPS62lQ
You can also run the Edge Impulse projects on the Pi easily. Just export to WebAssembly and you can call it from e.g. Node.js.
How is this a neural net? This process is so physical intense that errors are going to doom the process. You could spend years doing this and the resultant machine may be 30% accurate.