基于多摄像头网络和多尺度分析的组群与集群行为识别研究

项目名称： 基于多摄像头网络和多尺度分析的组群与集群行为识别研究

项目编号： No.61471235

项目类型： 面上项目

立项/批准年度： 2015

项目学科： 无线电电子学、电信技术

项目作者： 林巍峣

作者单位： 上海交通大学

项目金额： 80万元

中文摘要： 行为识别与分析技术在智能视频监控等应用中有着极为重要的作用,然而目前行为识别技术的研究仍然难以满足视频监控等应用的发展和需求。本项目拟结合申请人在行为识别和事件检测方面的前期研究工作以及模式识别和机器学习理论的发展趋势，分别从识别模型的构建和特征分析等方面对行为识别与分析领域中的几个重要问题进行重点研究并提出解决方法，主要体现为：(1) 提出基于网络传输模型的新方法，有效解决利用多摄像头实现联合行为识别的问题；(2) 提出新的多尺度特征分析及结构化模型，实现对组群中的各个尺度行为进行完整分析的目的；(3) 提出基于热模型的一致运动区域分割及行为分析等方法，突破高密集场景下对集群行为进行有效提取、分析、识别的难点。本项目的的研究成果将有力推动行为识别技术的研究及应用，为行为识别技术在智能视频监控等系统中的应用提供新的理论依据、关键技术、和可行算法。

中文关键词： 视频监控；多摄像头联合行为识别；多尺度行为分析；集群行为识别

英文摘要： Activity recognition and analysis is of increasing importance in many applications including video surveillance. However, the current development of activity recognition research is still unsufficient to satisfy the requirements of various video surveillance applications.In this project, we will investigate several important problems in activity recognition and activity analysis research. These important problems include the following: (1) We plan to address the multi- camera joint activity recognition problem by introducing a new network-based model to efficiently combine information from different cameras; (2) We plan to analyze the multi-scale activities for a group by performing multi-scale feature analysis and applying a structured model for recognition; (3) We plan to address the problem of crowd activity recognition in crowded scenarios by proposing a thermal-diffusion- based model which first segments the coherent motion regions from the scene and then performs crowd activity analysis accordingly. The research outputs of this project will greatly encourage the research and application of activity recognition techniques. Besides, this project can also provide new theorities and key algorithms to the applications of intelligent video surveillance.

英文关键词： Video Surveillance;Multi-camera-based Activity Recognition;Multi-scale Activity Analysis;Crowd Activity Recognition