| Project/Area Number |
17560352
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (C)
|
| Allocation Type | Single-year Grants |
|---|
| Section | 一般 |
|---|
| Research Field |
Communication/Network engineering
|
|---|
| Research Institution | Waseda University |
|---|
Principal Investigator |
MATSUMOTO Takashi Waseda University, Faculty of Science and Engineering, Professor (80063767)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
MURATA Noboru Waseda University, Faculty of Science and Engineering, Professor (60242038)
|
|---|
| Project Period (FY) |
2005 – 2007
|
| Project Status |
Completed (Fiscal Year 2007)
|
| Budget Amount *help |
¥2,980,000 (Direct Cost: ¥2,800,000、Indirect Cost: ¥180,000)
Fiscal Year 2007: ¥780,000 (Direct Cost: ¥600,000、Indirect Cost: ¥180,000)
Fiscal Year 2006: ¥600,000 (Direct Cost: ¥600,000)
Fiscal Year 2005: ¥1,600,000 (Direct Cost: ¥1,600,000)
|
|---|
| Keywords | online change detection / Bayesian framework / Sequential Monte Carlo Methods / non-linear feature extraction / marginal entropy minimization / Sequential Marginal Likelihood / 変化検出 / 粒子フィルタ |
|---|
| Research Abstract |
1. Online change detection is referred to as detecting abrupt changes behind give data online. Explicit functional form of the target behind given data is often unavailable. In addition, target system is often nonlinear so that linear algorithms are often unsatisfactory. This project proposed two online change detection algorithms within online Bayesian framework. One uses Sequential Marginal Likelihood while the other introduces a latent variable indicating abrupt changes. Both of them are implemented via Sequential Monte Carlo methods. The algorithms are verified with several examples.
2. Another algorithm is proposed in detecting faces in a video sequences. The algorithm is also based on an online Bayesian framework where it partly utilizes the Viola-Jones static score in a sequential manner. This is also implemented via Sequential Monte Carlo. It also checks Sequential Marginal Likelihood to detect any changes in which case the proposal distribution is reset. The algorithm is verified against real data.
3. In order to deal with complex features of data such as discontinuity of time series and deformation of images, a new non-linear feature extraction method is investigated based on the notion of marginal entropy minimization. The validity of the method is confirmed by neumerical experiments with real face data.
|
