| Summary: | 碩士 === 國立雲林科技大學 === 電子工程系 === 102 === Visual tracking has been one of the main research topics in computer vision for years. Although there are a variety of approaches developed for visual tracking, in this work we concentrate on the particle filter-based approach. Particle filter-based video object has achieved considerable success in recent years, enjoying the advantages of flexibility, ease of implementation, and capability to deal with nonlinear motion and non-Gaussian environments. However, tracking drift and failure may occur in certain complex situations. We adopt an incremental learning scheme for particle filter-based tracking, with the following modifications: (1) a new mechanism to select the optimal particle, and (2) online learning of the templates on a timely basis which copes with variations of the tracking environment. Extensive tests on real-life video sequences confirm the effectiveness of the proposed scheme.
|
|---|
