| 作成者 |
|
|
|
|
|
|
|
| 本文言語 |
|
| 発行日 |
|
| 収録物名 |
|
| 開始ページ |
|
| 終了ページ |
|
| 出版タイプ |
|
| アクセス権 |
|
| 関連DOI |
|
|
|
|
|
| 関連URI |
|
|
|
|
|
| 関連情報 |
|
|
|
|
|
| 概要 |
In this paper, we discuss a latency-free vision-based real-time human motion sensing system. Vision-based human motion sensing has a merit that it does not impose any physical restrictions on humans, ...which provides a natural way of measuring human motion. However, there are several issues to be solved. A topic discussed here is a computational aspect of visionbased human motion sensing. Since most of vision algorithms applied for human motion sensing are time consuming, latency, or delay, of the system, becomes large. This delay often becomes crucial especially to real time interactive applications. To solve this problem we introduce confidence-driven architecture, which has a mechanism of prediction, and which can make trade-off between latency and accuracy. In some cases, it is better to make the latency small even if the accuracy of the analysis becomes low, and, in other cases, it is required to get accurate results even if it is time consuming. There is a difficult trade-off between the latency and the accuracy, and it is required to control the tradeoff smoothly without rebuilding the system. In the confidence-driven architecture, the trade-off can be controlled by specifying a generalized parameter called confidence, which indicates how accurate the analysis should be. Applying this architecture to vision-based human motion sensing, we can develop a latency-free human motion sensing system.続きを見る
|
Mendeley出力
YoshimotoFCV04