基于卡尔曼预测粒子滤波的复杂场景图像目标跟踪算法  

Target Tracking Algorithm in Complex Scene Image Based on Kalman Predictive Particle Filter

作　　者：张博 ZHANG Bo(College of Information Science and Engineering,Changsha Normal University,Changsha 410100,China)

机构地区：[1]长沙师范学院信息科学与工程学院,长沙410100

出　　处：《自动化与仪表》2022年第6期74-78,共5页Automation & Instrumentation

基　　金：教育部中国高校产学研创新基金项目(2020ITA05028);教育部产学合作协同育人项目(201901014024);湖南省普通高等学校教学改革研究项目(HNJG-2021-1195);湖南省社会科学成果评审委员会一般项目(sskl202219)。

摘　　要：现有主流目标跟踪算法图像目标跟踪精度较差,无法满足现今复杂场景图像目标的跟踪,故该文提出基于卡尔曼预测粒子滤波的复杂场景图像目标跟踪算法研究。采用ViBe算法提取复杂场景图像中背景,以此为基础,构建图像目标状态转移模型与对应图像目标观测模型,确定目标运动参数的维度与粒子的权重,基于卡尔曼预测粒子滤波估计目标状态,通过粒子反复迭代与重采样,实现复杂场景图像目标的跟踪。实验数据显示,与7种主流算法相比较,提出算法RMSE数值最小,平均值为9.08;目标跟踪窗口较为规范,并且窗口内包含背景信息较少;在不同背景复杂度下,目标跟踪误差较小;在不同实验次数下,图像目标跟踪时间较短,平均值为10.30 ms,充分证实了提出算法应用效果较佳。The existing mainstream target tracking algorithms have poor image target tracking accuracy and cannot meet the current complex scene image target tracking.Therefore,a research on complex scene image target tracking algorithm based on Kalman predictive particle filter is proposed.The ViBe algorithm is used to extract the background in the complex scene image.Based on this,the image target state transition model and the corresponding image target observation model are constructed,the dimensions of the target motion parameters and the weight of the particles are determined,and the target state is estimated based on the Kalman prediction particle filter.Particles are repeatedly iterated and resampled to realize the tracking of complex scene image targets.The experimental data show that compared with the seven mainstream algorithms,the proposed algorithm has the smallest RMSE value,with an average value of 9.08.The target tracking window is relatively standardized,and the window contains less background information,under different background complexities,the target tracking error is smaller.Under different experiment times,the image target tracking time is short,and the average is 10.30 ms,which fully confirms that the proposed algorithm has better application effect.

关 键 词：复杂场景 图像 目标跟踪 实时性 卡尔曼预测粒子滤波 

分 类 号：TP399[自动化与计算机技术—计算机应用技术]