| 研究概要 |
Wearable computing is an important technology in health care. We developed a natural and light weight wearable interface to identify user gestures and activities for long-term life support. This year, we focused on the solutions of the gesture segmentation, recognition, and applications based on a finger-worn device, Magic Ring.
Segmentation of gesture sequence is a precondition in wearable computing. We proposed an adaptive threshold-based method based on Bayes Decision Theory and designed an adaptive mechanism to segment short-duration, ambiguous, and individually different gestures. The method can obtain high segmentation precision/recall and user-dependent adaptability with low computation complexity.
Recognition is the resolution of wearable computing. We proposed a pre-classification HMM method that can reduce recognition complexity by dividing gesture vocabularies into groups, maintain or even improve recognition accuracy by the adaptive adjustment of HMMs for different groups. Comparing with the one-fold HMM method, our method can greatly reduce computational complexity without any loss of recognition accuracy.
Application is the final goal of wearable computing. We employed only one finger-worn device to detect ten daily activities. To reflect realistic life aspects, a weight sequence alignment approach is proposed to analyze the detected activity sequences and attributes of each activity. The method can provide more detailed and realistic information of users' living for discover of potential health problem.
In summary, we developed a ring-shape device named Magic Ring that can detect finger/hand gestures and identify users' daily activities. The device has many potential applications like appliance control and health monitoring, which is especially useful for health care of elderly person in the aging society.
|
| 今後の研究の推進方策 |
In future, the research needs to be improved as least from the following aspects.
1) Device miniaturization. The finger-worn device will be re-designed to reduce the size to a real ring.
2) Data collection in real life. We will collect gesture data aiming to specific users in their real lives.
3) Improvement of segmentation and recognition method. Segmentation and recognition accuracy need to be improved to meet real application requirements in future. Especially for different user groups like young person and elderly person, the improved method based on the target users will be further studied.
4) Real Applications. Some core technologies have to be studied to meet real application requirements including real-time performance, energy saving, and competition avoid.
|
