Abstract: The understanding and recognition of human behavior is a key issue in an intelligence visual surveillance system. But most of the current research is on simple behavior with simple background, so it is not widely applicable. This article presents a method to model complex human behavior hierarchically in complex environment. Set some meaningful landmarks in the screen and model primitive behavior between two landmarks using a HMM, then use Level-Building algorithm with an appropriate threshold model and behavior grammar to recognize complex human behavior. Experimental results show that this method can achieve good results for both simple behavior and complex behavior in complex environment.

Key words: behavior recognition, HMM, Level-Building, behavior grammar

摘要： 人的行为模式的理解与识别是智能视觉监控系统的一个关键环节。针对目前大部分的研究都是简单场景下的简单行为识别，不具有广泛适用性的问题，该文提出一种复杂场景下的分层行为建模和识别方法。通过统计方法在监控画面内选定若干个有意义的标志点，利用这些标志点将复杂行为分解为一系列简单行为，对简单行为的轨迹进行HMM建模，并利用Level-Building算法进行复杂行为的识别。实验结果表明，该方法对复杂行为具有较高的识别率，而且在多种场景下具有普适性。

关键词: 行为识别, 隐马尔可夫模型, 分层构筑, 行为语法

CLC Number: 

• TP391.4