時系列濃度勾配ベクトル差分による微動人物領域抽出

概要

論文の詳細を見る
Extraction of human regions is an important research theme in the scene image analysis. The general conventional method to find human regions in a scene image is to evaluate the difference of gray-levels by pixels between the current frame and the registered background image without humans. However, chairs and books may be often extracted as the moved objects when a scene image in such a library is monitored. On the other hand, human regions cannot be extracted when they have few movements by the conventional method that evaluates the difference between the current frame and the previous frame. This paper proposes a method to extract human regions in a scene image by evaluating the differences of the density slope vectors at all pixels. The background image is not required for the proposed method. The experimental results show that the human regions were extracted successfully even if the persons sat down on the chairs without movements.
社団法人電気学会の論文
2005-06-01