基于视频流体模型的人体运动特征提取与运动过程语义建模

项目名称： 基于视频流体模型的人体运动特征提取与运动过程语义建模

项目编号： No.61262037

项目类型： 地区科学基金项目

立项/批准年度： 2013

项目学科： 自动化技术、计算机技术

项目作者： 唐权华

作者单位： 江西师范大学

项目金额： 43万元

中文摘要： (限400字)：利用单目视频进行人体运动识别是实现机器视觉、智能视频监控、基于内容视频检索等应用的基础。通过传统信号处理获得的视频特征与人体运动语义分离，性能依赖于训练样本，易受环境和人体随机运动干扰，致使人体运动识别准确率低。解决上述问题，发现语义相关的视频特征、建立基于语义特征、适应随机运动的人体运动模型，成为人体运动识别的关键问题之一。本项目立足于对单个人体运动的语义化描述的应用需求，基于视频流体模型研究人体运动的语义特征提取和运动过程建模方法。主要研究内容包括：(1)基于视频流纹的运动特征提取方法；(2)人体运动语义相关特征选择方法；(3)基于限时随机Petri网的人体运动过程与语义特征关系建模方法；(4)常见人体运动过程语义模型研究。通过研究人体运动过程建模方法，有助于实现更复杂、更灵活的视频内容理解，扩展人体运动识别的应用范围。

中文关键词： 运动识别；行为理解；视频流体模型；Petri网；

英文摘要： Human Motion Recognition (HMR) based on Monocular Video Sequences is a foundation of applications such as machine vision, Intelligent Video Surveillance, and content based video retrieval. The video features from traditional signal processing are separated from human motion semantic. Their performance of HMR depends on training samples, and is vulnerable to the environment and human random motion interference. To solve the problems above, the discovery of semantically related video features, and the human motion model based on semantic features to adapt to the random movement has become one of the key issues of HMR. Based on the application requirements of describing of a single human motion in the semantics, this application research the semantic feature extraction based on Video Flow Model and Modeling method of human motion process. The main contents include: (1) Video motion features extraction algorithm based on Video Flow Trace; (2) The human motion semantic feature selection algorithm; (3) Modeling method of Relation between human motion and semantic features based on Limited Probability Time Petri net; (4) Study of common human motion process semantic model. The study of human movement semantic modeling approach would help to achieve of more complex and flexible video content understanding would be benef

英文关键词： Motion Recognition；Activity Comprehension；Video Flow Model；Petri Net；