结合核相关滤波和Kalman预测的运动目标跟踪  

Moving Target Tracking Based on Kernelized Correlation Filter and Kalman Predicting

作　　者：田亚蕾 马杰 杨楠 TIAN Ya-lei;MA Jie;YANG Nan(School of Electronics and Information Engineering,Hebei University of Technology,Tianjin 300401,China)

机构地区：[1]河北工业大学电子信息工程学院,天津300401

出　　处：《小型微型计算机系统》2018年第10期2330-2334,共5页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61203245)资助;河北省自然科学基金项目(F2012202027)资助

摘　　要：核相关滤波(kernelized correlation filter,KCF)目标跟踪算法的跟踪过程容易受到快速运动和运动模糊、遮挡及相似物干扰等因素影响,导致跟踪效果不佳.针对以上问题,本文提出了一种结合KCF框架和Kalman预测器的运动目标跟踪算法.该算法首先将原始图像映射到CN(Color-Name)特征空间,使跟踪器能够处理多通道颜色特征的图像;其次在跟踪过程中,加入Kalman滤波器来预测图像目标位置,根据预测位置确定待检测区域进行检测,并利用检测结果更新Kalman滤波器,提高跟踪检测精度.在OTB-2013数据集上对改进的算法多次进行实验,并与其他6种较先进的跟踪算法进行对比,分析实验结果可知在目标发生快速运动、遮挡及相似物干扰等复杂情况下,本文方法均有较强的鲁棒性.The kemelized correlation filter （KCF} is easily interfered by occlusion, fast motion and motion blur, analog interference and other factors in the tracking process, which can result in tracking failure. Aiming at the above problems, this paper proposes a moving target tracking algorithm combining KCF framework and Kalman predictor. The algorithm first maps the original image to the CN （ color name ） feature space, which makes the KCF tracker process the image of the multi-channel color feature. Secondly, Kalman filter is added to predict the target position of the image during the tracking process. The area to be detected is determined according to the predicted position, and the Kalman filter is updated with the detection result to improve the tracking detection accuracy. The improved algorithm was tested several times on the OTB-2013 dataset and compared with the other six advanced tracking algo- rithms. Analyzing the experimental results, we can see that the proposed method is robust to deformation, occlusion, rotation, fast movement, and other complex scenes.

关 键 词：目标跟踪 核相关滤波 颜色特征空间 KALMAN滤波 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]