项目名称: 基于稀疏表示的多摄像机非重叠视野域运动目标跟踪方法研究
项目编号: No.61462052
项目类型: 地区科学基金项目
立项/批准年度: 2015
项目学科: 计算机科学学科
项目作者: 尚振宏
作者单位: 昆明理工大学
项目金额: 45万元
中文摘要: 由于目标出现在时空上离散,环境光照、摄像机成像特性差异以及跟踪过程中目标遮挡、外观变化等原因,使得多摄像机非重叠视野域跟踪成为具有挑战性的难题,该问题的解决对提高多摄像机监控系统的性能有重要意义。压缩感知理论的提出为解决多摄像机非重叠视野域跟踪提供了新的解决思路。本项目拟通过多特征融合构建目标高维稀疏特征,基于压缩感知理论对其降维,建立压缩感知域下多摄像机非重叠视野域运动目标特征模型;并通过将非重叠视野域下的目标关联问题转化为在关联数限制下的最大效用问题,建立最小费用流模型,一次求取所有关联数限制下的最大总体效用,实现多摄像机运动目标关联、跟踪。本项目以昆明市公安高清视频监控网络为实验和应用场景,以构建运动目标压缩感知域低维特征模型以及基于最小费用流的目标关联模型为突破口,为解决多摄像机非重叠视野域目标持续跟踪提供一些理论基础和新的技术手段。
中文关键词: 多摄像机;目标跟踪;稀疏表示;非重叠视野域
英文摘要: It is a challenging task to robustly track object in non-overlapping multi-camera networks due to factors such as discrete in time and space, illumination change, difference in cameras properties, occlusion. Solution of these problems is important to improve the performance of the multi-camera surveillance system. Theory of Compressive Sensing provides a new idea to resolve these problems. On the background of Kunming Peace City video surveillance, this project aims to propose a robust and consistent tracking method and some theory basis to resolve the challenging problem of tracking in non-overlapping multi-camera networks, based on the research of sparse features acquisition and representation of moving object, dimensionality reduction of high-dimensional sparse feature and moving object association,and a new object association model.
英文关键词: Muti-Camera;Object Tracking;Sparse Representation;Non-Overlapping Field View