摘要
针对复杂监控场景中往往无法准确检测前景目标,导致难以有效跟踪目标的问题,提出了一种基于属性关系图优化匹配的多运动目标跟踪方法,将目标跟踪问题转化为前景目标跟踪标记的优化问题,实现多运动目标跟踪。将前景区域按颜色、空间特征一致性划分为多个碎片;采用属性关系图描述目标模型,分析计算出属性关系图外观模型的属性相似度;提出通过概率松弛法计算目标函数并采用遗传算法进行优化匹配,得到前景碎片最优的跟踪标记,从而完成多目标跟踪。在多个监控视频上的实验结果表明,本方法能大大提高跟踪性能,实现复杂监控场景中的多目标跟踪。
It is difficult to effectively track multiple objects in complex scenes because the foreground objects can not be detected accurately. This paper proposes a novel algorithm for tracking multiple moving objects based on attributed relational graph optimizing match, and converts the problem of object tracking to the problem of foreground object tracking label optimization, and then realize multiple moving object tracking. Firstly, the foreground regions are divided into multiple patches according to color and spatial feature. The Attributed Relational Graph （ARG） is used to describe the appearance and structural feature of the object model, and the attribute similarity degree of ARG appearance model is analyzed and computed. The probability relaxation algorithm is used to calculate the object function, and the genetic algorithm is used iteratively to carry out optimization match, then the optimal tracking labels of the foreground patches are obtained, finally the multiple moving object tracking is realized. The tracking experiment results of multiple objects on surveillant videos indicate that the proposed approach can improve the tracking performanee greatly and realize multiple moving object tracking in complex surveillant scenes.
出处
《仪器仪表学报》
EI
CAS
CSCD
北大核心
2013年第3期608-613,共6页
Chinese Journal of Scientific Instrument
基金
国家863计划主题项目(2012AA112300)
国家自然科学基金项目(61005068)
湖南省自然科学基金项目(11JJ4049)
湖南省高校创新平台开放基金项目(11K019)
湖南省教育厅科研项目(10C0548)资助
关键词
视频监控
多目标跟踪
外观模型
属性关系图
概率松弛法
visual surveillance
tracking multiple object
appearance model
attributed relational graph
probability relaxation algorithm
