For many of us, pointing a device at an object and retrieving data about it has become part of our daily lives. The vast majority of our purchases will sport the ubiquitous barcode; an increasing number of printed magazine adverts, online articles and even television shows are using QR codes for access to more information; and most recently, near field communication technology is opening up new ways to interact with the world around us. A team of researchers from the Human-Computer Interaction Institute and Heinz College Center for the Future of Work Carnegie Mellon University has been looking into an alternative object tagging system called acoustic barcodes. The system takes the sound of a finger, pen or phone scraping across a series of parallel notches etched, embossed or cut into a surface or object, and converts it into a unique binary ID.
The barcode part of the system – developed by Chris Harrison, Robert Xiao and Scott Hudson – consists of a series of parallel grooves and ridges on the surface of an object designed to produce a unique, complex sound when something is scraped across the top, such as a fingernail or the edge of a smartphone. Each notch is between 0.25 and 0.5 mm thick, 0.1 to 0.3 mm deep and 7 mm wide, and they "are separated by a small integer multiple of a unit gap distance, either 1.6 or 3.2 mm."
A guard sequence made up of three notches separated by unit gaps sits at the beginning and end of each payload and the system provides for two different binary code schemes. In the fixed-notch-count scheme, a groove followed by a one-unit-length gap is resolved as a zero and a groove followed by two unit lengths translates to a one. The fixed-physical-length scheme, however, encodes each one as a notch and each zero as the space in between.
The team has already applied the grooves to many different objects and materials, including wood, glass, paper, acrylic, stone and polyester transparencies, and says that although not tested, the acoustic barcode should also work on metal.
The short burst of sound produced as a fingernail or phone body is swiped across the barcode is picked up by an inexpensive piezo contact microphone attached to a surface or object. The microphone used for the team's experiments is reported to have cost just US$6 and is capable of monitoring a surface area of 10 square meters (107.6 sq ft).
Recording begins if the system detects input above a predetermined level, so as not to misfire and record the neighborhood dogs barking instead of a swipe across the barcode. The sound is sampled at 96 kHz, preprocessed with a high-pass-filter at 4 kHz to remove background hum and any human speech from the signal, and then the waveform is cleaned up and smoothed out to accentuate the peaks and subdue the constants.
It's then fed into a peak detection routine and onto a barcode decoder which resolves the acoustic barcode as a unique binary ID.
The reading system has also been developed to compensate for variations in swipe speed by calculating a "unit length implied by each gap (the gap length divided by the unit multiple). This is then averaged with the previous unit estimates, allowing the value to drift as decoding proceeds."
The team has also confirmed that the system is capable of using the internal microphone of mobile devices to register the acoustic barcodes.
The system has been tested for accuracy by a small user group who swiped six different types of barcodes using three devices – a fingernail, a dry erase marker and a mobile phone. The microphone was attached to the whiteboard for the first two test conditions and to the mobile phone itself for the latter. To avoid problems associated with varying swipe distances, the fixed-physical-length encoding scheme was used for this part of the study.
Each test user swiped five different bars once with each of the three devices for a total of 270 trials over the course of the experiment, which were processed in real time.
At the close of the evaluation, it was found that system accuracy was a little less than perfect, and varied depending on which device was swiped across the barcode. The smartphone performed best (87.4 percent accuracy), followed by the fingernail (77.9 percent) and then the marker (66.4 percent).
The researchers point out that any real-world application of this kind of system would need to incorporate some form of error correction mechanism for improved robustness. However, it has the potential to be used like traditional product barcodes to retrieve information, for example, or augment the display on a smartphone screen, start apps or trigger functions.
Source: Chris Harrison
You can see the system outlined in the video below, along with a few suggested applications.
