交通视频中车辆行为理解与异常事件感知关键技术研究

项目名称： 交通视频中车辆行为理解与异常事件感知关键技术研究

项目编号： No.61272350

项目类型： 面上项目

立项/批准年度： 2013

项目学科： 自动化技术、计算机技术

项目作者： 熊璋

作者单位： 北京航空航天大学

项目金额： 80万元

中文摘要： 在海量交通视频中开展车辆行为理解和异常事件感知方法研究对于解决交通拥挤、确保运输安全具有重要的意义。针对复杂的环境因素和车辆行为不确定性的影响，本课题通过民主融合策略建立空间-颜色混合模型，并利用图割理论实现对遮挡的处理，实现高鲁棒性、抗遮挡的车辆跟踪方法。通过对车辆轨迹聚类方法的改进，获得车辆的静态和动态特征，利用基于MGHMM的半监督车辆行为模型和语义区域模型实现车辆行为的细分与理解。主要研究内容包括：1) 基于多视觉特征融合的多车辆目标跟踪方法；2）基于图割的多车辆目标跟踪遮挡处理方法；3）基于LWPGMA规则的车辆轨迹聚类算法；4）半监督车辆行为模型的优化与车辆行为的细分。通过在关键技术研究中的探索和创新，完成对车辆行为的感知、理解和异常事件的检测，并在此基础上，实现一个交通视频中车辆行为理解与异常事件感知原型系统，验证方法和技术的有效性和实用性。

中文关键词： 智能交通系统；车辆跟踪；车辆轨迹聚类；车辆行为理解；异常事件感知

英文摘要： Over the massive traffic surveillance videos, the research on vehicle behavior understanding and abnormal events perception has decisive significance for settling traffic congestion, guaranteeing traffic safety. According to the analysis on the complicated environmental factors and the uncertainty of vehicles behavior, the robust and anti-block multiple-vehicle tracking method is presented by setting up the SMOG model with democratic integration strategies and tracking multiple objects through occlusion using the bi-label graph cut algorithm. The static and dynamic characteristics of multiple-vehicle can be extracted with the improved vehicle trajectory clustering method. Based on the semi-supervised adaptive detection model with MGHMM and semantic region, a novel segmentation and understanding method of vehicle behaviors is proposed. The outstanding items involved are as follows: 1) adaptive integration of multiple visual features for vehicle tracking; 2) vehicles segmentation and tracking under occlusions using extended Graph-Cut with Bi-Label; 3) a novel trajectory cluster method based on length weighted pair-group method using arithmetic averages (LWPGMA); 4) the segmentation and optimization of semi-supervised adaptive vehicle events model. On the basis of these studies, the prototype system of vehicle beha

英文关键词： intelligent transportation system；vehicle tracking；vehicle trajectory cluster；vehicle behavior understanding；Abnormal Vehicle Events Perception