Micah Bojrab
Published © LGPL

Low-Power Image Classification

The project combines machine learning with low-power hardware to enable ultra portable devices to understand their surrounding world.

AdvancedShowcase (no instructions)Over 6 days2,396
Low-Power Image Classification

Things used in this project

Hardware components

Intel Edison
Intel Edison
×1
Intel Edison Arduino Breakout Board
This should include a
×1
Microsoft LifeCam HD-3000
×1
Plugable USB 2.0 10-Port Hub
I found this on Amazon for cheap. The brand is "Plugable." This needs to be externally powered.
×1
9V Battery Clip
9V Battery Clip
×1
9V battery (generic)
9V battery (generic)
×1
SanDisk 8GB MicroSD card
×1
US Coin Currency
×1

Story

Read more

Schematics

Configuration

This is an image showing the overall connections. The usb hub is used to power most usb devices. The webcam can also be connected through the hub if desired.
Connecting the Edison to battery is important when powering devices, such as the camera, because it needs to be in Host mode.
There is a microSD card connected to the slot on the Arduino Breakout Board, which provided extra space needed to install the software and trained networks. This needs to be reformatted to EXT3 in order to install Anaconda, because of the symbolic links associated with the installation.
The small micro-usb A cord in the far right can connect to a laptop to provide the serial connection for a terminal.
Img 4650

Code

NN Libraries built on Theano

These are the libraries and tools associated with collecting, cleaning, preparing and training the imagery. I chose US coins for the initial implementation as it has lower dimensionality than other classifications. Included are some pre-trained networks, an example training dataset used for the competition, and the final applications to classify and count the total value of the coins captured by the webcam.

Credits

Micah Bojrab

Micah Bojrab

1 project • 9 followers
Software Engineer going back to school for a PhD, and specializing in Computer Vision and Machine Learning.

Comments