面向多目标视频跟踪的出生强度估计方法  

Birth Intensity Estimation Method for Multi-target Video Tracking

作　　者：周小龙[1] 沈恒华[2] 何炳蔚[2] 

机构地区：[1]浙江工业大学计算机科学与技术学院,杭州310023 [2]福州大学机械工程及自动化学院,福州350108

出　　处：《计算机辅助设计与图形学学报》2014年第12期2223-2231,共9页Journal of Computer-Aided Design & Computer Graphics

基　　金：国家自然科学基金(61273286;51175087);福建省杰出青年基金(2013J06013)

摘　　要：针对多目标视频跟踪中的新生目标出生强度估计问题,提出一种有效的基于熵分布和覆盖率的方法.该方法利用前一时刻所获目标状态及测量值对出生强度进行初始化,再利用当前时刻所获测量值对出生强度进行更新.在更新阶段,首先选取仅依赖于权值的负指数熵分布作为出生强度的先验分布,滤除出生强度中与当前时刻测量值无关的噪声分量;然后通过计算出生强度与相应测量值间的覆盖率对出生强度权值进行再次更新,进一步滤除权值小于给定阈值的噪声分量.实验结果表明,文中方法有效地降低了噪声分量的影响,提高了多目标视频跟踪的准确率.In this paper, an effective birth intensity estimation method that based on entropy distribution and coverage rate is proposed for multi-target video tracking. The birth intensity is first initialized according to the previously obtained target states and measurements. The currently obtained measurements are then used to update the initialized birth intensity. In the updating stage, the negative exponent entropy distribution that depends on the weight is first selected as the prior distribution of the birth intensity. By doing so, the components within the birth intensity those are irrelevant to the measurements could be regarded as noises and should be removed. The coverage rate between each birth intensity component and the corresponding measurement is then computed to further eliminate those components whose weights are smaller than the given threshold. Experiments on noisy video sequences are conducted to show that the proposed birth intensity estimation method can effectively eliminate the noises and finally improve the tracking accuracy.

关 键 词：多目标视频跟踪 高斯混合概率假设密度滤波器 出生强度估计 熵分布 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]