项目名称: 基于可变拓扑模型的多姿态行人检测研究
项目编号: No.61473086
项目类型: 面上项目
立项/批准年度: 2015
项目学科: 其他
项目作者: 杨万扣
作者单位: 东南大学
项目金额: 84万元
中文摘要: 智能视频监控能为社会的发展提供安全保障,其对于社会公共安全建设具有越来越重要的作用。行人检测是智能视频监控的关键步骤,也是当前模式识别领域的研究热点和难点,姿态变化是行人检测面临的重要挑战之一。本课题的研究内容主要围绕行人检测中的多姿态问题进行深入研究:(1)引入压缩感知理论、VC 维最小化、相关性最小化等概念设计分类器以构建部位检测器;(2)建立基于多元混合高斯模型的人体拓扑结构;(3)研究基于隐参数的模型优化方法;(4)构建基于卷积神经网络的拓扑深度学习模型。本课题的主要成果要形成发明专利;同时还要在国际权威期刊和主流的国际学术会议上发表一系列高质量的学术论文;通过本项目的研究,将显著推动行人检测研究,提升智能视频监控安防功能。
中文关键词: 行人检测;拓扑模型;压缩感知;分类器设计;深度学习
英文摘要: Video surveillance is one of most important element for society safety, and Government gave more and more attention on the practical applications of object detection. Pedestrian detection is a hot research topic in pattern recognition and is much valuable in theory and application. This project mainly focuses on the variation of posture in pedestrian detection. In this project, we will do the following work: (1)design classiers based on compressed sensing theory, VC-dimension minimization and relevance minimization for parts detectors. (2)construct topological models using multi variants gaussian models; (3) explore the latent parameters classifier optimization theory; (4) consturct topological models based deep learning. The project will significantly promote the research in pedestrian detection and improve the performance of pedestrian detection in video surveillance.
英文关键词: pedestrian detection;topological model;compressive sensing;classifier design;deep learning