摘要
为了在背景干扰、遮挡环境下可靠地跟踪目标,提出了一种特征融合式粒子滤波跟踪算法,改造了传统的边缘方向直方图模型,使其对平移更敏感,以利于提高跟踪精度;利用均值漂移模块检测干扰,依据两种特征对目标和干扰的区分能力调节它们在观测模型中的权重,以抑制干扰;设计混合相似度指标检测遮挡,实时调整系统模型以适应遮挡环境;通过在模型更新过程中引入相互监督机制,缓解模型漂移问题,实验结果表明,算法是有效的。
To track object in the case of background distractor and occlusion, a particle filter based tracker was proposed. Traditional edge direction histogram is reconstructed to become more sensitive to translation with the aim of improving tracking accuracy. Mean shift module is used to detect distractor. Weight of each histogram models can be adjusted according to its ability to distinguish the object and dis- tractor. Hybrid similarity measurement is designed to detect occlusion, and the detecting result is used to adjust system model. Interactive supervision Mechanism is introduced into the model updating process to alleviate the problem of model drift. Experimental results show the effectiveness of the algorithm.
出处
《计算机测量与控制》
CSCD
北大核心
2009年第11期2292-2294,共3页
Computer Measurement &Control
基金
国家自然科学基金重点项目(60634030)
高等学校博士学科点专项科研基金资助(20060699032)
关键词
目标跟踪
粒子滤波
特征融合
遮挡处理
object tracking
particle filter
feature fusion
occlusion handling
作者简介
左军毅（1975-），男，西安人，工学博士，主要从事多目标跟踪理论及应用、信息融合、飞行器制导与控制等方向的研究。
