基于上下文协作、多级观测和数据关联的复杂场景多目标跟踪

项目名称： 基于上下文协作、多级观测和数据关联的复杂场景多目标跟踪

项目编号： No.61305011

项目类型： 青年科学基金项目

立项/批准年度： 2014

项目学科： 自动化技术、计算机技术

项目作者： 路红

作者单位： 南京工程学院

项目金额： 25万元

中文摘要： 对基于视觉的多目标跟踪(MTT)研究是智能视频监控系统的重要研究内容。融合目标外观和场景上下文的协作跟踪机制用于MTT问题的研究中，可有效突破复杂场景下部分遮挡、全遮挡、相似特征目标干扰、持续外观变化等问题，并使对MTT鲁棒、准确和实时性问题的解决得到改善。本项目研究内容为：(1)利用目标外观局部和全局共生特征分析、状态建模，研究空时域目标外观上下文观测建模和更新；(2)从目标级和协作级两个层面，分析目标、目标间全局和局部状态参数一致性，研究多目标检测、关联信息挖掘及协作表征理论；(3)从"目标组"级、目标级和协作级三个层次，研究多目标数据关联、观测融合和跟踪优化。项目研究目标为：力求提出一套有效的适用于复杂场景MTT的目标外观和场景上下文挖掘和表征方法，并在此基础上提出多目标检测和数据关联方法，为建立更有效解决MTT鲁棒、准确和实时性间平衡关系的理论和方法，为MTT实用化提供理论支持。

中文关键词： 多目标跟踪；上下文协作；多级观测；数据关联；复杂场景

英文摘要： Vision-based multi-target tracking (MTT) is a significant research topic in intelligent surveillance system. Collaborating track mechanism that combines target appearance and scene contexts proves an effective way in MTT system to handle partial and total occlusions, remove disturbances induced by similar targets, manage continuous target appearance changes under complex scenarios etc in that it improves the robustness, accurateness and real-time processing of the tracking system. Taking into account all these factors, this project will focus on the following three aspects. Firstly, we will analyze the part-whole co-occurrence feature and model the target state to stuy spatio-temporal target appearance context modeling and updating. Secondly, from the object and collaboration aspects, we will utilize the part-whole state parameters coincidences of the target itself and among the targets to study multi-target detection, correlative information mining and collaborating description. Thirdly，from the target group, target and collaboration levels, we will investigate multi-target data association, observation fusion and tracker optimizing. The project aims to propose a set of effective approaches to target appearance and scene contexts mining and description of the MTT under complex scenarios, and then put forward mu

英文关键词： Multi-target tracking；context collaboration；multi-level observation；data associations；complex scenarios