Activity Recognition from On-Body Sensors: Accuracy-Power Trade-Off by Dynamic Sensor Selection
Activity recognition from an on-body sensor network enables context-aware applications in wearable computing. A guaranteed classification accuracy is desirable while optimizing power consumption to ensure the system’s wearability. In this paper, we invest
- PDF / 615,983 Bytes
- 17 Pages / 430 x 660 pts Page_size
- 15 Downloads / 221 Views
epartment of Electronic Informatic and System, University of Bologna, Italy {pzappi,efarella,lbenini}@deis.unibo.it www.micrel.deis.unibo.it 2 Wearable Computing Lab., ETH Z¨ urich, Switzerland {lombriser,stiefmeier,droggen,troster}@ife.ee.ethz.ch www.wearable.ethz.ch
Abstract. Activity recognition from an on-body sensor network enables context-aware applications in wearable computing. A guaranteed classification accuracy is desirable while optimizing power consumption to ensure the system’s wearability. In this paper, we investigate the benefits of dynamic sensor selection in order to use efficiently available energy while achieving a desired activity recognition accuracy. For this purpose we introduce and characterize an activity recognition method with an underlying run-time sensor selection scheme. The system relies on a meta-classifier that fuses the information of classifiers operating on individual sensors. Sensors are selected according to their contribution to classification accuracy as assessed during system training. We test this system by recognizing manipulative activities of assembly-line workers in a car production environment. Results show that the system’s lifetime can be significantly extended while keeping high recognition accuracies. We discuss how this approach can be implemented in a dynamic sensor network by using the context-recognition framework Titan that we are developing for dynamic and heterogeneous sensor networks.
1
Introduction
Wearable computing aims at supporting people by delivering context-aware services [1]. Gestures and activities are an important aspect of the user’s context. Ideally they are detected from unobtrusive wearable sensors. Gesture recognition has applications in human computer interfaces [2], or in the support of impaired people [3]. Developments in microelectronics and wireless communication enable the design of small and low-power wireless sensors nodes [4]. Although these nodes have limited memory and computational power, and may have robustness or accuracy limitations [5,6], unobtrusive context sensing can be achieved by integrating them in garments [7,8] or accessories [9]. R. Verdone (Ed.): EWSN 2008, LNCS 4913, pp. 17–33, 2008. c Springer-Verlag Berlin Heidelberg 2008
18
P. Zappi et al.
In an activity recognition system, high classification accuracy is usually desired. This implies the use of a large number of sensors distributed over the body, depending on the activities to detect. At the same time a wearable system must be unobtrusive and operate during long periods of time. This implies minimizing sensor size, and especially energy consumption since battery technology tends to be a limiting factor in miniaturization [10]. Energy use may be reduced by improved wireless protocols [11,12], careful hardware selection [13], or duty cycling to keep the hardware in a low-power state most of the time [14]. Energy harvesting techniques may also complement battery power [15], although the unpredictability of energy supply typical of harvesting makes it difficult to m
Data Loading...
