AIY

With all the buzz in AI, I have taken the initiative to begin exploring aspects of it as part of my self-directed 'Making Moments'. Those are the moments when I get inspired to make, tinker and innovate. One of those moments led me to getting for myself a Google AIY (Artificial-Intelligence-Yourself) Vision Kit.

Google's AIY Vision Kit Joy Detector feature gif from programmersought.com

It is not exactly cheap but for a Do-It-Yourself Aritifical Intelligence Maker Kit (Yes, AIY is a play on DIY), it is worth it as it definitely provides a hands-on introduction to AI and machine learning concepts. According to the AIY site by Google,...

"The AIY Vision Kit from Google lets you build your own intelligent camera that can see and recognize objects using machine learning. All of this fits in a handy little cardboard cube, powered by a Raspberry Pi.

Everything you need is provided in the kit, including the Raspberry Pi."

Also, once I got my AIY kit delivered to my home, I was already working from home and the circuitbreaker here kept me mostly indoors so I began my attempts at putting together the AIY Vision. The instructions online were fairly clear and straightforward and I was able to put together the pieces eventually. 

So these there are some features of the AIY Vision Kit worth noting from this site I came across in designnews.com:

There's a Raspberry Pi Zero WH at the Core.

The Google AIY kits are powered by a Raspberry Pi Zero Wireless with a Header single board computer. The main processor is a BCM 2835 SoC with an operating speed of 1 GHz. The BCM has 512MB of RAM, which is the same storage capacity as the original Raspberry Pi A+. Included on the Pi Zero WH is a BCM43143 processor with built in WiFi and Bluetooth Low Energy (BLE) features. The board’s computing and wireless electronic components are populated on a 65 x 31 x 11.6-mm (2.6” x 1.2” x 0.5”) PCB. The Raspberry Pi Zero WH with all its electronic components and features weighs 11.5g (0.4oz).

The Google AIY Vision Kit Uses a Convolutional Neural Network.

Image recognition allows machines to identify objects, places, people, and text in their surrounding environment. This is typically done with algorithms like optical character, pattern or gradient matching, face or scene identification, or scene detection techniques.

The Google AIY Vision kit’s image recognition feature has been enhanced with a convolutional neural network (CNN). A CNN can support an abundant number of neurons, thereby expressing large models through computations. Each layer of the AIY vision kit’s CNN allows higher level, more abstract features of an object to be detected. The AIY vision kit uses a 2D CNN to recognize images and a 3D CNN to recognize images and colors of an object.

A Vision Processing Unit (VPU) is used for object detection, recognition, and classification.

The AIY vision kit uses a vision bonnet for object detection and image recognition through classification for camera enhancement. An Intel Movidius VPU is onboard the electronic bonnet for assisting in the object detection and image recognition-classification and processing event. The VPU can use a variety of image sensors for deep CNN classification of detected objects. The processing unit has gesture/eye tracking and recognition capabilities along with 3D depth detection. There is an inertial measurement unit (IMU) externally wired to the VPU. The IMU assists the AIY vision kit in detecting camera orientation for object clarity and recognition. "

Here are some photos I took in my unboxing and making process. Unfortunately, there is no unboxing video but you can see the progress of the intelligent camera as I took lots of photos along the way. 

 

Putting together everything went fine. The next step was to make the camera work and then possibly consider how different machine learning algorithms can be introduced (through the SD card) and how the functions can come in handy in educational technology. As a start the joy detector could be a demo to the time when tech can detect students' emotions accurately and thus provide timely data for teachers to act upon if needed. That is more on the affect side, but there are potentially other ways that this Vision technology can benefit Teaching and Learning (T&L). I aam still exploring and troubleshooting some of these and will share more in this blog when I make some meaningful headway.