基于OpenCV的组合优化多目标检测追踪算法
|
摘要
为了改善计算机视觉领域中多目标检测与追踪算法的实时性,综合考虑实现多目标跟踪过程的两个主要阶段:目标检测与目标追踪。第一阶段采用改进的自适应高斯混合背景建模及滤波处理方法,通过间隔性选取的图像帧作为高斯混合背景模型更新方式,进而从视频图像中检测出明确有效的多个目标的同时兼顾算法实时性,此阶段仿真结果显示该算法使得效率提高40%左右;在目标追踪阶段采用一种快速高斯粒子滤波追踪算法,使用按照高斯分布的初始化粒子的线性变换来近似拟合高斯粒子分布,从而在原算法无需粒子重采样的基础上避免了粒子的高斯拟合过程,进一步提高运算效率,增强实时性。最后以测试视频为研究对象结合上述两种优化算法进行仿真,结果显示此组合优化方法可使运算速率提高将近30%。
In order to improve the instantaneity of the multi-object detection and tracking algorithm in the computer vision field, this paper considered the two main phases of the multi-object tracking process: object detection and object tracking. In the first stage, an improved adaptive Gaussian mixture background model and a smoothing processing method were adopted to extract clear and effective targets from the image frame alternatively, and the method' s instantaneouity was taken into account.The simulation results show that the operation efficiency increase about 40%. In the object tracking phase, a fast Gaussian particle filter tracking algorithm was adopted. The Gaussian particle distribution was approximated by linear transformation of initialized Gaussian particles. The Gaussian fitting process was avoided on the basis of the original algorithm without particle resampling, to further improve the operation efficiency and enhance the instantaneity performance. Finally, a test video was used to test on the combined optimization method, the results show that this method improves the speed by about 30%.
In order to improve the instantaneity of the multi-object detection and tracking algorithm in the computer vision field, this paper considered the two main phases of the multi-object tracking process: object detection and object tracking. In the first stage, an improved adaptive Gaussian mixture background model and a smoothing processing method were adopted to extract clear and effective targets from the image frame alternatively, and the method' s instantaneouity was taken into account.The simulation results show that the operation efficiency increase about 40%. In the object tracking phase, a fast Gaussian particle filter tracking algorithm was adopted. The Gaussian particle distribution was approximated by linear transformation of initialized Gaussian particles. The Gaussian fitting process was avoided on the basis of the original algorithm without particle resampling, to further improve the operation efficiency and enhance the instantaneity performance. Finally, a test video was used to test on the combined optimization method, the results show that this method improves the speed by about 30%.
引文
[1]ZIVKOVIC Z.Improved adaptive Gaussian mixture model for background subtraction[C]//ICPR 2004:Proceedings of the 17th International Conference on Pattern Recognition.Washington,DC:IEEE Computer Society,2004:28-31.
[2]JAFFER A J,BAR-SHALOM Y.On optimal tracking in multiple target environments[C]//NOLTA 1972:Proceedings of the Third Symposium on Non-Linear Estimation Theory and Its Applications.Piscataway:IEEE,1972:112-117.
[3]KOTCHA J H,DJURIC P M.Gaussian particle filtering[J].IEEE Transactions on Signal Processing,2003,51(10):2592-2601.
[4]KOTECHA J H,DJURIC P M.Gaussian sum particle filtering[J].IEEE Transactions on Signal Processing,2003,51(10):2602-2612.
[5]尹宏鹏,陈波,柴毅,等.基于视觉的目标检测与跟踪综述[J].自动化学报,2016,42(10):1466-1489.
[6]张红颖,李鸿,孙毅刚.基于混合高斯模型的阴影去除算法[J].计算机应用,2013,33(1):31-34.
[7]王典.基于混合高斯的背景建模与阴影抑制算法研究[D].西安:西北工业大学,2006:14-20.
[8]YANG H X,SHAO L,ZHENG F,et al.Recent advances and trends in visual tracking:a review[J].Neurocomputing,2011,74(18):3823-3831.
[9]李天成,范红旗,孙树栋.粒子滤波理论、方法及其在多目标追踪中的应用[J].自动化学报,2016,41(12):1981-2002.
[10]CREAL D.A survey of sequential Monte Carlo methods for economics and finance[J].Econometric Reviews,2012,31(3):245-296.
[11]陈鹏,钱徽,朱淼良.一种快速高斯粒子滤波算法[J].华中科技大学学报(自然科学版),2008,38(增刊1):291-294.
[12]JIA Z,LI Y,MAO M,et al.Research of improved probability data association algorithm for multi-target tracking[C]//CCDC2009:Proceedings of the 21st Annual International Conference on Chinese Control and Decision Conference.Piscataway:IEEE,2009:4955-4959.
[13]KRISTAN M,MATAS J,LEONARDIS A,et al.The visual object tracking VOT2015 challenge results[C]//ICCV 2015:Proceedings of the 2015 International Conference on Computer Vision Workshops.Piscataway:IEEE,2015:564-586.
[2]JAFFER A J,BAR-SHALOM Y.On optimal tracking in multiple target environments[C]//NOLTA 1972:Proceedings of the Third Symposium on Non-Linear Estimation Theory and Its Applications.Piscataway:IEEE,1972:112-117.
[3]KOTCHA J H,DJURIC P M.Gaussian particle filtering[J].IEEE Transactions on Signal Processing,2003,51(10):2592-2601.
[4]KOTECHA J H,DJURIC P M.Gaussian sum particle filtering[J].IEEE Transactions on Signal Processing,2003,51(10):2602-2612.
[5]尹宏鹏,陈波,柴毅,等.基于视觉的目标检测与跟踪综述[J].自动化学报,2016,42(10):1466-1489.
[6]张红颖,李鸿,孙毅刚.基于混合高斯模型的阴影去除算法[J].计算机应用,2013,33(1):31-34.
[7]王典.基于混合高斯的背景建模与阴影抑制算法研究[D].西安:西北工业大学,2006:14-20.
[8]YANG H X,SHAO L,ZHENG F,et al.Recent advances and trends in visual tracking:a review[J].Neurocomputing,2011,74(18):3823-3831.
[9]李天成,范红旗,孙树栋.粒子滤波理论、方法及其在多目标追踪中的应用[J].自动化学报,2016,41(12):1981-2002.
[10]CREAL D.A survey of sequential Monte Carlo methods for economics and finance[J].Econometric Reviews,2012,31(3):245-296.
[11]陈鹏,钱徽,朱淼良.一种快速高斯粒子滤波算法[J].华中科技大学学报(自然科学版),2008,38(增刊1):291-294.
[12]JIA Z,LI Y,MAO M,et al.Research of improved probability data association algorithm for multi-target tracking[C]//CCDC2009:Proceedings of the 21st Annual International Conference on Chinese Control and Decision Conference.Piscataway:IEEE,2009:4955-4959.
[13]KRISTAN M,MATAS J,LEONARDIS A,et al.The visual object tracking VOT2015 challenge results[C]//ICCV 2015:Proceedings of the 2015 International Conference on Computer Vision Workshops.Piscataway:IEEE,2015:564-586.