基于光流和概率上下文无关文法的人行为分析

摘要

文中提出了一个基于光流和概率上下文无关文法的人的视频序列的运动行为的分析方法。首先，采用自适应混合高斯模型来提取背景，然后采用形态学方法确定目标前景的轮廓。使用光流确定视频序列中的关键帧，并对关键帧的前景轮廓进行概率上下文无关文法分析。%This paper presents a method of human behavior analysis based on optical flow and probabilistic context free grammar. First, an adaptive mixture Gaussian background model is used to extract background pixels, and the foreground contours are obtained by morphology methods. The optical flow is used to obtain key frames in video sequences, and the foreground contours of key frames is analyzed by probabilistic context free grammar.