Adapting Magnetic Resonant Coupling Based Relative Positioning Technology for Wearable Activitiy Recogniton
Abstract
We demonstrate how modulated magnetic field technologythat is well established in high precision, stationary motiontracking systems can be adapted to wearable activity recognition.To this end we describe the design and implementationof a cheap (components cost about 20 Euro for the transmitterand 15 Euro for the receiver), low power (17mA forthe transmitter and 40mA for the receiver), and easily wearable(the main size constraint are the coils which are about25mm3) system for tracking the relative position and orientationof body parts. We evaluate our system on two recognitiontasks. On a set of 6 subtle nutrition related gestures it achieves99.25% recognition rate compared to 94.1% for a XSense inertialdevice ( operated calibrated, euler angle mode). On therecognition of 8 Tai Chi moves it reaches 94 % compared to86% of an accelerometer. Combining our sensor with the accelerometerleads to 100% correct recognition (as comparedto 90% when combining the accelerometer with a gyro).
1 Introduction
Posture and motion of body parts are well known to be akey component of many human activities. Therefore, a significantproportion of research in activity recognition is basedon posture and motion information. However, technology forcapturing body motion and posture (often referred to as motiontracking systems) is still far from being perfect. A muchcited overview of such technologies ([9]) is titled ”No silverbullet but a respectable arsenal”. From wearable applicationpoint of view the problem is that many of the best technologiesin the ’respectable arsenal’ require stationary, often bulkyinfrastructure and are not suitable for wearable use. Today,the core of wearable activity work still relies on accelerometerswhich capture only a small part of posture and motioninformation (see Related Work below).General Idea Existing magnetic field systems (e.g. the”flock of birds” system from ascension: ascensiontech.com) require a bulky, power consuming stationary transmitterand cost thousands of dollars. In our system both thetransmitter and the receiver consist of a couple of cheap components(price of all components 20 Euro for the transmitterand 15 Euro for the receiver) consume a reasonable amount ofpower (17mA for the transmitter and 40mA for the receiver),and are easily wearable (the main size constraint are the coilswhich are around 25mm3).Our work is based on the observation that the requirementsthe system must fulfill to be a useful tool for activityrecognition are very different from the requirements forwhich stationary magnetic systems have been built. Stationarymagnetic systems target sub centimeter precision, high speedtracking over ranges of up to 3 meter. On the other hand, foractivity recognition, the system merely needs some sort of relativeposition related signal over short distances (50 to 80cm).The signals do not have to be the actual position and orientationvalues expressed in any standard units. They just need tobe deterministic and reproducible.Paper Organization The paper first describes the physicalprinciple behind the system and outlines and justifies the assumptionsbehind it. It then provides a detailed descriptionof the implemented system including the discussion of differentdesign alternatives and problems. We finish with a threestep evaluation of the system. First we conduct an analysisof the signals produced by our system. Second we comparethe performance on a simple gesture classification task againstthe Xsens MT9 inertial tracking sensors. There our systemsgets 99.25% recognition rate compared to 94.1% for a singleMT9 and 97.28% for a system of 3 MT9 sensors. Finally wedemonstrate the benefit of combining our sensor with an accelerometeron a more complex activity recognition task: theclassification of Tai Chi moves. Starting with a recognitionrate of 86% for the accelerometer, and 94% for our systemthe combination of the two leads to 100% correct recognition(accelerometer/gyro combination 90%).Related Work The use of accelerometers for motion basedactivity recognition is motivated by the availability of cheap,low power, easy to handle sensors. The disadvantage of accelerometersis that they are able to capture only a smallpart of motion and posture information


Download full report
http://doi.ieeecomputersociety10.1109/ISWC.2008.4911584