Tikalon Header

Pattern Recognition for Accelerometers

December 9, 2013

At the first major energy crisis in 1979, people were willing to explore commuter transportation modes other than the personal automobile. The corporate headquarters campus at which worked at that time contained both business offices and research laboratories and a huge number of employees. The parking lots were as huge as those of many shopping malls, and desk employees got some needed exercise walking long distances from their cars to their offices.

Whether conscious to the needs of its employees, or spurred into action by some governmental mandate, my employer started a partially-subsidized commuter van pool fleet. There were nearly twenty vans at its peak, each van capable of transporting twelve-fifteen employees from nearby suburbs to their workplace. The drivers were the employees, themselves, who would rotate the driving task in some vans, while in other vans there was a designated employee driver who got a free ride.

There were benefits beyond my chauffeured ride to work. We scientists in the lab are mostly shielded from the business operations of a corporation, but in the van I had the enjoyment of hearing how the other half worked. At the same time, my fellow commuters got an education in some of the science happening in nearby buildings.

Fantasy car pool, circa 1943.Fantasy carpool, circa 1943, featuring two men and eight women.

From a World War II poster promoting fuel conservation.

(Office of War Information poster, via Wikimedia Commons.)

The only downside was a somewhat bumpy ride to work. New Jersey has a moderately intense winter season, and the multiple freeze-thaw events take their toll on the roads (If you know New Jersey, you'll know why that's an intended pun). If I sat at the back of our van, the vibration assured that I would be fully-awake when I arrived at work. I often thought about bringing an accelerometer along for the ride to see exactly how bad the bumps were. From accelerometer data, a van ride to work could be distinguished from an automobile ride.

As its name implies, an accelerometer is a sensor for measuring acceleration. There are high accuracy accelerometers for inertial navigation applications, but today's MEMS technology makes it very easy to manufacture inexpensive, moderate accuracy accelerometers. They are so inexpensive that smartphones and tablet computers use them to select page orientation.

Accelerometers in mobile computing devices can be used, also, for other applications. I haven't checked, but there must be a few applications designed to have your cellphone say, "Ouch," when it's dropped on the floor. A multi-megapixel camera is not that useful when it's hand held, but an accelerometer can be used for image stabilization.

Scientists at the Berkeley Seismic Laboratory have proposed using mobile device accelerometers as an earthquake early warning system. Accelerometers would transmit data to a central server which would use an algorithm to sort seismic tremors from the random motions of individual devices and transmit an alarm. At this time, mobile device accelerometers can only detect earthquakes above 5.0 magnitude, but they might contain more sensitive accelerometers in the future.[1]

Computer scientists at the University of Helsinki have reported on their system of using mobile device accelerometers to determine commuter transportation habits.[2-5] Finland is a hotbed of computer science, enabled by its excellent Internet connectivity,[6] and users of the Linux operating system might recognize the University of Helsinki as the alma mater of Linus Torvalds, first developer and proponent of the Linux kernel. Torvalds is still active in Linux development; and, yes, Linux is named after him.

Linus Torvalds in September, 2012Linus Torvalds, first developer of the Linux operating system, at a meeting in Tianjin, China, September 11, 2012

(World Economic Forum photograph, via Wikimedia Commons.)

The Helsinki team has verified my informal hypothesis that vibration monitoring can distinguish one mode of transportation from another, and they've summarized their system for this, called TMD-Peaks, in a paper for the 11th ACM Conference on Embedded Networked Sensor Systems (SenSys'13, Rome, Italy).[3] Project leader, Samuli Hemminki, mentions the same random background problem faced by the Berkeley seismic sensor team,
"Extracting vehicular movement information from smartphone accelerometers is challenging as the placement of the device can vary, users interact with the phone spontaneously, and as the orientation of the phone can change dynamically. We overcame these challenges by developing novel algorithms for processing and analyzing accelerometer measurements."[3]
Some typical data are shown in the following figure. Experiments have shown that the mode of transportation can be determined with 80% accuracy.[3] The team is working on an Android implementation.[2]

vehicle accelerometer dataVehicle accelerometer data.

Top to bottom: train, bus, stationary, metro, tram, automobile.

(University of Helsinki image, reformated for presentation.)

References:

  1. Jonathan Amos, "Smartphones to be pocket seismometers," BBC News, December 5, 2012.
  2. TMD-Peaks: Accelerometer-Based Transportation Mode Detection on Smartphones, University of Helsinki Web Site.
  3. Smartphone accelerometers distinguish between different motorized transportation modalities, University of Helsinki Press Release, November 13, 2013 (PDF file).
  4. S. Hemminki, P. Nurmi and S. Tarkoma, "Accelerometer-Based Transportation Mode Detection on Smartphones," Proceedings of ACM SenSys, 2013 (In Press, PDF File).
  5. TMD-Peaks Presentation Slides (PDF File).
  6. Top 50 Countries With The Highest Internet Penetration Rate, Internet World Stats. At the time of this article, Finland was ranked tenth best in the world with 88.6% of the population having access, with the US lagging at 27th place, with 78.3% of the population having access.