| Summary: | 碩士 === 國立東華大學 === 資訊工程學系 === 98 === Several foreground segmentation and shadow removal algorithms have been proposed for simple or only special cases. We develop a Modified List-Scheduling Model (LSM) and utilize a Motion History Image (MHI) to perform foreground segmentation. We try to address the problem of dynamic background with frequent movements like the waving tree/grass, or even transient changes like the switching of projecting slides in a classroom. The kernel concept of MLSM is to maintain a list of chain structure for each pixel to record the history information of the appearance features including color, texture, and motion. According to the appearance frequencies and feature variations, the order of each list will be dynamically adjusted to update the status of each background pixel. Finally, we combine multiple features to handle the dynamic background to improve the robustness of the foreground detection. The proposed system is implemented in an e-Learning system called VEC3D (Virtual English Classroom in 3D).
|
|---|
