复杂背景下MeanShift结合Kalman滤波的车辆跟踪算法

详细信息    查看全文 | 推荐本文 |

英文篇名：Vehicle tracking algorithm based on MeanShift combined with Kalman filter in complex background 作者：苏灵松 英文作者：Su Lingsong;Institute of Optical Electrical and Computer Engineering,University of Shanghai for Science and Technology; 关键词：复杂背景 ; 车辆跟踪 ; MeanShift ; Kalman滤波 英文关键词：complex background;;vehicle tracking;;MeanShift;;Kalman filtering 中文刊名：DZCL 英文刊名：Electronic Measurement Technology 机构：上海理工大学光电信息与计算机工程学院; 出版日期：2018-03-23 出版单位：电子测量技术 年：2018 期：v.41;No.290 语种：中文; 页：DZCL201806013 页数：5 CN：06 ISSN：11-2175/TN 分类号：77-81

摘要

针对复杂背景下车辆跟踪的难点,提出一种新的跟踪算法来更好地实现复杂背景下的车辆跟踪。该算法融合MeanShift和Kalman两种跟踪方法各自的优点,即在MeanShift跟踪过程中插入Kalman预测车辆下一步的运动位置。首先,通过设置好Kalman滤波功能选择的阈值,对MeanShift的跟踪结果进行判断,若MeanShift跟踪结果理想,则Kalman滤波的功能是平滑跟踪结果;若跟踪结果不理想,则Kalman滤波的功能是预测下一帧车辆的位置。最终试验结果说明该算法能有效提升跟踪精度和鲁棒性,实时性提高了17%左右,车辆丢失率降低了20%~30%左右,能更好地针对复杂背景下完成对车辆的跟踪。
        Aiming at the difficulties of vehicle tracking in complex background,a new tracking algorithm is proposed to better realize the vehicle tracking in complex background.The algorithm combines the advantages of MeanShift combined Kalman's two tracking methods.That is,during the MeanShift tracking process,insert the Kalman to predict the next movement of the vehicle.In our method,we combined the MeanShift and Kalman tracking methods.That meant Kalman filter was added to the vehicle during the MeanShift process to predict the vehicle.First,the tracking results of MeanShift are judged by setting the threshold selected by the Kalman filter function.If the MeanShift tracking results are ideal,then the Kalman filter function is smooth tracking results,otherwise Kalman filter function is to predict the next frame of the vehicle position.The results of the final experiments showed the real-time increased by about 17%,so this method can effectively improve the tracking accuracy and robustness.And the vehicle loss rate decreased by 20% to 30%.It can be better for the complex background to complete the tracking of the vehicle.

引文

[1]唐雨操.基于车辆识别的车辆追踪与计数[J].信息与电脑(理论版),2016(13):97-98.
    [2]周越,张帆,夏彩凤.一种基于改进型MeanShift算法的运动目标跟踪方法[J].科技经济导刊,2017(21):184.
    [3]朱闻亚.结合稳健估计和MeanShift的视频目标跟踪算法[J].沈阳工业大学学报,2017,39(2):177-182.
    [4]龙涛.基于Kalman预测器改进的CamShift目标跟踪[D].湘潭:湘潭大学,2016.
    [5]葛如海,张雪峰,张美娟.基于PreScan的侧后方车辆识别与跟踪算法的研究[J].汽车技术,2017(8):32-37.
    [6]程武.基于OpenCV运动目标检测与跟踪方法研究[D].沈阳:沈阳航空航天大学,2017.
    [7]乔国娜.基于CamShift与混合高斯模型的目标跟踪与提取算法[J].计算机与数字工程,2016,44(6):1053-1056.
    [8]聂义.智能视频监控中单目标运动的检测与跟踪算法研究[D].长沙:湖南师范大学,2016.
    [9]陈菲,黄勇坚,李淑彦,等.自适应窗口的MeanShift目标跟踪算法[J].电子技术,2016,45(5):1-4.
    [10]唐徵羽.基于低空无人机平台的运动车辆检测跟踪系统[D].深圳:深圳大学,2017.
    [11]DUAN H,ZHANG Q,DENG Y,et al.Biologically eagle-eye-based autonomous aerial refueling for unmanned aerial vehicles[J].Chinese Journal of Scientific Instrument,2014,35(7):1450-1457.
    [12]MORANDUZZO T,MELGANI F.Automatic car counting method for unmanned aerial vehicle images[J].IEEE Transactions on Geoscience and Remote Sensing,2013,52(3):1635-1647.
    [13]朱钾,李化.一种基于MeanShift算法的目标跟踪改进[J].中北大学学报(自然科学版),2016,37(6):633-637.
    [14]刘苗,黄朝兵.基于多特征自适应融合的MeanShift目标跟踪方法[J].现代计算机(专业版),2016(11):68-72.
    [15]王相海,刘美瑶,苏元贺,等.基于Contourlet和MeanShift的交通视频车辆跟踪算法[J].辽宁师范大学学报(自然科学版),2017,40(2):192-198.
    [16]陈双叶,王善喜.基于五帧差分和改进的MeanShift算法的运动目标跟踪[J].计算机科学,2016,43(S1):203-206.