多摄像机人体姿态跟踪  

Human posture tracking in a multi-camera environment

作　　者：孙洛[1] 邸慧军[1] 陶霖密[1] 徐光祐[1] 

机构地区：[1]清华大学计算机科学与技术系,普适计算教育部重点实验室,北京100084

出　　处：《清华大学学报（自然科学版）》2011年第7期966-971,共6页Journal of Tsinghua University(Science and Technology)

基　　金：国家自然科学基金资助项目(60673189,60433030)

摘　　要：人体姿态跟踪是人类行为分析和识别的基础,具有广阔的应用范围。该文以人体的头肩部轮廓构成的Ω形状的曲线作为跟踪目标,利用多摄像机间的位置约束和形状约束,提出了一种基于多摄像机的人体朝向、位置和头肩部轮廓的联合跟踪算法。该算法将多视角头肩部轮廓跟踪和三维空间中的头肩部位置跟踪、朝向检测融合于同一框架中。真实场景实验表明该算法具有良好的鲁棒性和准确性。Human posture tracking has long been of interest for its wide applications in human motion recognition and analysis.A tracking algorithm was developed using the Ω-shaped curve of the human head and shoulder as the tracking target based on the position and shape constraints among distinct views in a multi-camera environment.The algorithm integrates tracking and detection to simultaneously track the human posture and location in 3-D space as well as the head-shoulder contour.The robustness and efficacy of the approach is confirmed in challenging real-world scenarios.

关 键 词：姿态跟踪 多摄像机 粒子滤波器 动态Bayes网 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]