9 of 10Image
Teaching computers to recognise objects is getting even easier, thanks to devices like the Arduino Pixy.
The Pixy is a fast vision sensor that can be taught to find objects in the real world and report its findings via simple interfaces. The device is a small camera, about half the size of a business card, attached to an Arduino board that handles the image recognition.
Teaching Pixy to identify objects requires users to place an item in front of the Pixy and press the button on top of the device. Pixy then generates a statistical model of the colours of the object that it stores in flash and later uses to identify the item.
Software in the Pixy detects objects using a hue-based colour filtering algorithm. Pixy calculates the hue and saturation of each RGB pixel from the image sensor and uses these as the primary filtering parameters for detecting objects. The hue of an object remains largely unchanged with changes in lighting and exposure, making it an effective way to detect objects.
The makers of Pixy claim it can identify hundreds of objects in a scene at a time, using its connected components algorithm, and then report back on each object's size and location through one of its interfaces.
The device consists of an Omnivision OV9715 0.25–inch image sensor with 1280x800 resolution and an NXP LPX4330 microcontroller with a dual-core ARM processor, which can process images at 50 FPS. The Pixy has interfaces for UART serial, SPI, 12C, digital and analogue I/O.
An application called PixyMon allows users to see what Pixy sees, outputting either raw or processed video, and allowing users to configure Pixy, set the output port and manage color signatures.
The Pixy is a joint development between Carnegie Mellon University and Texas-based Charmed Labs.
The Kickstarter campaign for Pixy raised more than ten times its target of £25,000 and the first 3,500 Pixy cameras are expected to ship on 14 January next year. The camera is still available to order for $75.00, including shipping.
While we wait for washing machines that Tweet at us when the laundry's done or homes that text you to let you know the basement's flooded there's Twine.
Twine is pocket-sized box of sensors with a wireless internet connection built in, that can be configured to keep you updated about what's happening in the world.
Twine is configurable using a simple web app, which allows you to set rules about when and in what situation Twine should message you, for example 'WHEN'+'moisture sensor is wet' then 'SEND'+text message: 'The basement's flooded'. Messages can be sent via Twitter, email, text message, or as straight HTTP GET or POST requests to feed data into a web app. Once Twine has been configured it functions as a standalone device.
The box has sensors to detect its temperature, orientation and vibration, and can be expanded with a moisture sensor, external thermometer, a magnetic switch (which could be used for detecting if a door is ajar), and an Arduino Shield or breakout board to work with more complex electronics projects.
Twine is small, just 2.7 inches square in size, and can keep running for months on two AAA batteries.
On Kickstarter the project raised more than $555,000, far more than the $35,000 originally asked for. Twine is available to buy now for $124 through the project website.
3Doodler is a pen that squirts out melted plastic rather than ink, allowing users to draw 3D models in the air or on a surface.
Plastic is pushed out through a 0.3mm nib, which heats the plastic to make it pliable so it can be pushed out of the pen, before hardening into a solid object as it cools.
The pen has been used to create a variety of objects, including a 60cm Eiffel Tower and a detailed model of a hand. 3Doodler plans to release different sized tips next year.
The 3Doodler is now shipping to its Kickstarter backers and is available to order at the 3Doodler website for $99. The project attempted to raise $30,000 through Kickstarter, but ended up raising $2.3m.