摘要
针对复杂背景下的矿井跟踪视场由于单一线索对目标缺乏可分性、观测模型可靠度对场景变化缺乏自适应性致使观测失效、跟踪发散的问题,提出了基于自适应多测量融合UPF的矿井人员跟踪算法.采用UKF产生预测样本,通过融入最新观测的建议分布引导预测样本分布在状态空间的高似然区域,扩大了预测样本与观测似然峰值重叠区域;提出运动光流直方图,将其与颜色直方图融合建立多观测模型,根据贡献率度量因子实现对观测模型可靠度的动态调节,定义了采样补偿函数有效克服观测失效时的粒子扩散.结果表明:本算法能够有效的解决矿井跟踪视场下(背景复杂)由于观测模型失效而导致的跟踪发散问题,与3种不同观测模型PF算法做状态估计MSE比较,估计准确率提高87%.
To resolve the target-tracking in coal mines,which was caused by a single-cue lacking of discrimination to target features and the reliability of observation modals lacking of self-adaptation to changes of scenes,a novel unscented particle filter(UPF) algorithm was proposed for object-tracking using an unscented kalman filter(UKF)as proposal distribution and an adaptive multi-cues fusion modal as the observation modal.To integrate the latest observation into proposal distribution,the UKF generate prediction samples were adjusted to high likelihood area in state-space,and more overlap regions of prediction samples and peak zones of observation likelihood were achieved.The optical flow histogram was presented and observation model based on multi-cues fusion was implemented by integrating optical flow with color.The adaptive strategy of observation modal weights was implemented by adjusting the contribution rate of single-cue observation modal,and the reliability of an observation modal was adjusted.A function of sample compensation was proposed to handle particle diffusion due to failure of observation modal.The results show that the tracking algorithm is effective for solving tracking failure in coal mines(complex background).The estimating accuracy was increased by 87% compared with three other particle filter algorithms.
出处
《中国矿业大学学报》
EI
CAS
CSCD
北大核心
2011年第1期146-151,共6页
Journal of China University of Mining & Technology
基金
国家高技术研究发展计划(863)项目(2008AA062200)
江苏省产学研联合创新基金项目(BY2009114)
关键词
UKF
粒子滤波
运动光流直方图
观测权值自适应
采样补偿
unscented kalman filter
particle filter
moving optical flow histogram
adaptive weights of observation modal
resample compensation