摘要
提出的基于三维在线表观模型的粒子滤波目标跟踪算法,以目标的独立特征为基础,分别从空域和时域对目标进行描述,构建目标的三维表观模型,并通过多重线性空间理论表达目标表观随时间推移引起的变化,实现模型的在线增量更新。采用粒子滤波方法,对每个独立线索分别进行在线权重估计,通过多线索的融合实现动目标的稳定跟踪。三维在线表观模型和在线跟踪机制使跟踪模型对目标与背景的在线区分能力得到进一步增强,保证了算法在目标表观变化时的跟踪稳定性。通过多种目标表观复杂变化的场景验证,均取得了良好跟踪效果。
In actual particle filter tracking, the appearance change of objects tends to be very changeful. To address this problem, an adaptive object tracking algorithm based on three-dimensional on-line appear- ance model is presented. It describes objects by spatial and temporal domains and constructs object＇s three-dimensional appearance model. Indicating object appearance changes over time by multiple linear space elements, it implements the on-line incremental update of models. Based on the tracking mechanism of particle filter, it estimates and weights each individual element cue online and tracks real-time adaptive moving targets. Three-dimensional online appearance model and online tracking mechanism make the dis- tinguish ability improved between the object and background. Therefore, it ensures the stability of the tracking algorithm when the object appearance changes. Experimental results show that the proposed tracking method could accurately track the moving object on a variety of challenging sequences, and it demonstrates better stability compared with related algorithms, while the target apperance presents vari- ous complex changes.
出处
《光电子．激光》
EI
CAS
CSCD
北大核心
2015年第9期1768-1775,共8页
Journal of Optoelectronics·Laser
基金
国家自然科学基金(F0111)
陕西省自然科学基金(2013JQ8023)资助项目
关键词
目标跟踪
表观模型
线索融合
在线跟踪
线性空间
object tracking
appearance model
multi-cue fusion~ on-line tracking~ linear space
