基于优化M-S模型的多目标鲁棒跟踪  

Robust multi-target tracking method based on an improved Mumford-Shah model

作　　者：苏洁[1,2] 印桂生[1] 魏振华[3] 刘亚辉[4] 

机构地区：[1]哈尔滨工程大学计算机科学与技术学院,黑龙江哈尔滨150001 [2]哈尔滨理工大学计算机科学与技术学院,黑龙江哈尔滨150080 [3]华北电力大学控制与计算机工程学院,北京102206 [4]北京信息科技大学计算中心,北京100085

出　　处：《哈尔滨工程大学学报》2010年第9期1228-1233,共6页Journal of Harbin Engineering University

基　　金：国家自然科学基金资助项目(60775058);教育部科学技术研究重点资助项目(107028)

摘　　要：针对光照变化情况下多遮挡目标的跟踪准确率差的问题,提出了一种基于优化M-S模型的鲁棒多目标跟踪算法.利用抗噪声性能高的优化M-S模型实现复杂环境下多目标精确识别与提取,降低模糊边缘、噪声的影响;利用区域像素标记方法建立目标和背景的边缘特征,在目标发生相互遮挡情况下也能够提取各个目标独立、完备的边缘特征.为了降低联合粒子滤波的计算复杂度,提高跟踪实时性,提出了简化联合滤波跟踪模型.仿真实验证明了该算法的正确性和有效性,与经典的差分跟踪算法、基于颜色特征的跟踪算法比较,对噪声边缘和变化光照环境敏感性降低,跟踪有效率统计分析表明鲁棒性提高1.82%,准确率提高1.36%.To improve the poor multi-target tracking accuracy that has resulted when images with variations in illumination and blocking of targets are processed,a robust multi-target tracking method based on an improved Mumford-Shah model was proposed.The optimized Mumford-Shall model has high noise immunity,and it was used to improve identification and extraction accuracy for multi-targets in complex environments.It was also able to reduce the effects of blurred edges and noise.Methods of pixel marking were used to record the target regions and back-ground regions.By doing so,independent and complete edge-feature sets could be determined,even when the targets shielded each other.A simplified combined multi-filter tracking model was proposed to reduce computational complexity and improve real-time tracking.Correctness and accuracy were tested through experiments.Compared to the standard difference-based tracking algorithm and color-based tracking algorithms,sensitivity to blurred edges and variations in surrounding illumination when tracking were reduced.Statistical analysis of tracking efficiency showed that robustness improved by 1.82% and accuracy improved by 1.36%.

关 键 词：多目标跟踪 M-S模型 边缘特征 水平集 联合滤波 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]