稀疏化、结构化和判别性约束的多视角行为识别方法研究

项目名称： 稀疏化、结构化和判别性约束的多视角行为识别方法研究

项目编号： No.61202168

项目类型： 青年科学基金项目

立项/批准年度： 2013

项目学科： 计算机科学学科

项目作者： 高赞

作者单位： 天津理工大学

项目金额： 24万元

中文摘要： 自动地多视角人体行为识别已成为智能监控、人机交互和数字娱乐等领域亟待解决的问题。为了实现多视角环境下人体行为的准确识别，拟提出稀疏化、结构化和判别性约束的多视角人体行为建模方法。该方法包括目标检测和跟踪，特征提取和多视角行为建模三部分，本课题将侧重解决多约束的多视角行为建模问题，其主要内容包括四个方面：首先，模型构建：拟通过多视角问题下多约束条件的挖掘和相应正则项的制定来构建多视角行为识别目标函数；其次，词典学习：拟基于协同表示和Fisher准则实现多视角联合词典学习；再次，针对多视角信息存在较强相关性，拟基于最大熵准则实现多视角联合词典冗余信息去除；最后，模型推断：拟基于坐标下降法实现多参数模型联合最优推断。本课题研究有利于多视角人体行为识别系统构建，相关研究成果对计算机视觉和模式识别研究领域新问题和新方法的探索具有重要意义，并为相关工业界新应用的发展提供可行和创新性技术基础。

中文关键词： 行为识别；多视角；稀疏表示；词典学习；

英文摘要： Automated multi-view human behavior recognition is essential for intelligent surveillance, human-machine interface, digital entertainment and etc. In this proposal, we propose the multi-view human behavior recognition method induced by sparsity, structuring and discrimination constraints. The proposed method consists of three steps, object detection and tracking, visual feature extraction, and multi-view behavior modeling. Our research will focus on the third step. Specifically, it includes four main problems: 1) Problem Formulation: designing the objective function for this task by discovering the constraints for multi-view requirement and formulating the corresponding regularization terms; 2) Dictionary Learning: learning the dictionary with multi-view information based on collective representation and Fisher Discriminant Criterion; 3) Redundancy Elimination: reducing the redundant bases in the learned dictionary with Maximum Entropy Criterion; 4) Model Inference: predicting the optimal parameters based on coordinate decent method. This research will promote the system construction for multi-view human behavior recognition. Moreover, it will not only benefit the research of computer vision and pattern recognition but also provide the feasible and novel techniques for multiple applications.

英文关键词： behavior recognition；multi-view；sparse representation；dictionary learning；