Research on Action Recognition Model Human Behavior Understanding

本研究的目的是实现动作识别模型，这是人类行为理解的基础技术。人类行为的理解是人类共生系统或人类共生机器人的必要条件。本研究利用了具有高速视觉跟踪能力的视觉处理系统。本研究提出了三层层次模型。最上层是行为歧视。其次是动作的注意，最低的是动作的跟踪。通过实验验证了该模型的有效性，该模型采用了基于人体行为模型库的人体部位发现/跟踪过程、多注意区域处理和行为识别过程。该模型用于两种类型的应用。一是对医疗室的医疗了解。大多数医学耳鼻喉治疗都可以使用该模型和构建的系统来理解。另一个应用是对躺在床上的病人的指向行为（指向手势）的理解。所提出的模型用于理解一个人的手势。在该应用中，颜色特征跟踪显示出在复杂环境下的鲁棒性。

The purpose of this research is to realize a action recognition model which is a basic technology for human behavior understanding. Human behavior understanding is necessary for human symbiosis systems or human symbiotic robots. This research utilizes visual processing system which has high-speed visual tracking capability.This research proposed three-revel hierarchy model. The top level is discrimination of behavior. The second is attention of action and the lowest is tracking of motion. The model is verified by expeiments which uses, human parts finding/tracking process, multiple attention region processing, and behavior discrimination process based on human behavior model-base.The model has utilized for two types application. One is for Medical treatment understanding in medical care room. Most of medical ear/nose treatment can be understood using the proposed model with constructed system. The other application is pointing behavior (pointing gesture) understanding of sick person in a bed. The proposed model is used to understand one's gesture. In this application, color feature tracking revealed to be robust in complicated environment.