基于人工鱼群算法的动态目标跟踪技术研究
|
摘要
粒子滤波方法是近年提出的一种适用于目标跟踪的有效算法,但存在粒子的退化现象,导致许多状态更新的估计对目标跟踪轨迹不起任何作用,在浪费大量计算资源的同时降低了粒子滤波器的性能。因此研究一种实时性,准确性和鲁棒性好的寻优方法是目标跟踪的关键问题。人工鱼群算法具有快速跟踪极值点漂移的能力,及较强的跳出局部极值点的能力,因此在运动目标跟踪领域具有重要的应用价值。本文对基于人工鱼群算法的目标跟踪系统进行研究。
本文在分析了人工鱼群算法和粒子滤波算法理论基础上,构建人工鱼群目标跟踪模式,该模式采用人工鱼的觅食、聚群、追尾三种行为策略,不同的行为模式定义粒子不同的状态转移方程,根据试探行为接近目标程度而选不同的状态转移模型,避免了粒子滤波算法相同的状态转移模式,使粒子具有智能群体的行为。仿真实验表明,人工鱼群算法在动态目标速度和方向骤变,以及存在遮挡物情况下,避免目标丢失现象,为解决动态目标跟踪提供了新方法。
本文利用进化规划的变异算子,增加人工鱼的多样性,扩大鱼群的分布搜索范围,加快获得全局最优解的速度。采用协同控制技术,构建多个子人工鱼群的分布体系,对每一个子群体进行三种行为试探,将最优子群体的状态指导其它群体的移动,实现了个体与群体的信息交互与相互协作机制,消除人工鱼漫无日的随机游动。仿真实验表明该方法能够适应目标运动的随机性和动态性,扩大了搜索范围,提高跟踪性能。
The particle filter method is an effective algorithm which is put forward for target tracking. Due to the degeneration phenomenon in the algorithm of particle filter, state estimate to a trajectory may be invalid, resource is wasted, and performance is reduced. Therefore, the method with real-time, accuracy and robustness is a key for the target tracking. The abilities of the maximum drifting and jumping out of a local extreme value are possessed by the algorithm of artificial fish swarm. So, the algorithm of artificial fish swarm has an important application value in the area of target tracking. Target tracking using the algorithm of artificial fish swarm is studied in the paper.
Based on the analysis of the algorithms of artificial fish and particle filter, the target tracking model by artificial fish swarm is proposed in the paper. The behaviors of preying, clustering, tailgating in artificial fish swarm are adopted. Different behavior is defined as different state transition formula which is selected to use according to the distance near a target. Thus, the same transition formula in classical particle filter is avoided and this makes particles more intelligent. Simulation results show that the algorithm can be used to solve the losing target phenomenon when a motion target speeding and changing direction suddenly or being blocked. A new method is proposed in target tracking model.
In order to expand distribution scope and quickly get a global optimal state, the evolution of mutation operator is used to increase the diversity of artificial fish in the paper. Based on the technology of synergistic control, artificial fish is divided into different groups in distribution system. In different groups three behaviors are tested and the better value is gained which is used to guide the motion of other groups. The cooperation and mutual information between any individual and groups are realized. At the same time, the aimless random swimming of artificial fish is eliminated. Simulation results show that the target moving randomly and dynamically can be tracked, the search scope expanded, and tracking performance improved by the proposed method.
本文在分析了人工鱼群算法和粒子滤波算法理论基础上,构建人工鱼群目标跟踪模式,该模式采用人工鱼的觅食、聚群、追尾三种行为策略,不同的行为模式定义粒子不同的状态转移方程,根据试探行为接近目标程度而选不同的状态转移模型,避免了粒子滤波算法相同的状态转移模式,使粒子具有智能群体的行为。仿真实验表明,人工鱼群算法在动态目标速度和方向骤变,以及存在遮挡物情况下,避免目标丢失现象,为解决动态目标跟踪提供了新方法。
本文利用进化规划的变异算子,增加人工鱼的多样性,扩大鱼群的分布搜索范围,加快获得全局最优解的速度。采用协同控制技术,构建多个子人工鱼群的分布体系,对每一个子群体进行三种行为试探,将最优子群体的状态指导其它群体的移动,实现了个体与群体的信息交互与相互协作机制,消除人工鱼漫无日的随机游动。仿真实验表明该方法能够适应目标运动的随机性和动态性,扩大了搜索范围,提高跟踪性能。
The particle filter method is an effective algorithm which is put forward for target tracking. Due to the degeneration phenomenon in the algorithm of particle filter, state estimate to a trajectory may be invalid, resource is wasted, and performance is reduced. Therefore, the method with real-time, accuracy and robustness is a key for the target tracking. The abilities of the maximum drifting and jumping out of a local extreme value are possessed by the algorithm of artificial fish swarm. So, the algorithm of artificial fish swarm has an important application value in the area of target tracking. Target tracking using the algorithm of artificial fish swarm is studied in the paper.
Based on the analysis of the algorithms of artificial fish and particle filter, the target tracking model by artificial fish swarm is proposed in the paper. The behaviors of preying, clustering, tailgating in artificial fish swarm are adopted. Different behavior is defined as different state transition formula which is selected to use according to the distance near a target. Thus, the same transition formula in classical particle filter is avoided and this makes particles more intelligent. Simulation results show that the algorithm can be used to solve the losing target phenomenon when a motion target speeding and changing direction suddenly or being blocked. A new method is proposed in target tracking model.
In order to expand distribution scope and quickly get a global optimal state, the evolution of mutation operator is used to increase the diversity of artificial fish in the paper. Based on the technology of synergistic control, artificial fish is divided into different groups in distribution system. In different groups three behaviors are tested and the better value is gained which is used to guide the motion of other groups. The cooperation and mutual information between any individual and groups are realized. At the same time, the aimless random swimming of artificial fish is eliminated. Simulation results show that the target moving randomly and dynamically can be tracked, the search scope expanded, and tracking performance improved by the proposed method.
引文
[1]李晓磊.一种新型的智能优化算法——人工鱼群算法[D].杭州：浙江大学,2003.
[2]李晓磊,邵之江,钱积新.一种基于动物自治体的寻优模式：鱼群算法[J].系统工程理论与实践,2002(11).
[3]工闯.人工鱼群算法的分析及改进[D].大连：大连海事大学,2008.
[4]李晓磊,钱积新.基于分解协调的人工鱼群优化算法研究[J].电路与系统学报,2003(1).
[5]窦全胜,周春光,徐中宇等.群体启发进化规划[J].计算机科学,2005(7).
[6]栾丹.改进的进化规划算法[J].渤海大学学报(自然科学版),2005(2).
[7]Kramer K A, Stubberud S C. Tracking of multiple target types with a single neural extended kalman filter[J]. Intelligent Systems,2006(1).
[8]Khairnar Dilip Gopichand, Merchant S N. Nonlinear target identification and tracking using UKF[J]. Granular Computing,2007(1).
[9]袁泽剑,郑南宁,贾新春.高斯-厄米特粒子滤波器[J].电子学报,2003(7).
[10]Gordan N, Salmond D, Smith A. Novel approach to nonlinear/non-Gaussian Bayesian state estimation[J]. Proceedings on Radar and Signal Processing,1993(1).
[11]Amaud D, Simon G, Chistophe A. On sequential Monte Carlo sampling methods for Bayesian filtering[J]. Statistics and Computing,2000(10).
[12]Freitas J F G, Niranjan M, Gee A H. Sequential Monte Carlo methods to train neural network models [J]. Neural Computation,2000(4).
[13]K Murphy, S Russell. Rao-Blackwellised particle filtering for dynamic Bayesian networks[C]. New York:sequential Monte Carlo methods in practice,2001.
[14]Cheng C, Rashid A, Ashfaq K. Multiple object tracking with kernel particle filter[C]. 2005(1).
[15]Yang ShuYing, Ge WeiMin, Zhang Cheng. Tracking unknown moving targets on omnidirectional vision[J]. Vision Research,2009(3).
[16]徐宗本,张讲社,郑亚林.计算智能中的仿生学：理论与算法[M].北京：科学出版社,2003：9-20.
[17]Xiaojuan Shan, Mingyan Jiang, Jingpeng Li. The routing optimization based on improved artificial fish swarm algorithm[C]. Intelligent Control and Automation, 2006 (1).
[18]李亮,迟世春,林皋.禁忌鱼群算法及其在边坡稳定分析中的应用[J].工程力学,2006(3).
[19]张梅凤.人工鱼群智能优化算法的改进及应用研究[D].大连：大连理工大学,2007.
[20]黄光球,王西邓,刘冠.基于网格划分策略的改进人工鱼群算法[J].微电了学与计算机, 2007(7).
[21]高德芳,赵勇,郭杨.基于混合鱼群一蚁群算法的模块化产品配置设计[J].设计与研究,2007(1).
[22]宋志字,李俊杰,汪红宇.混沌人工鱼群算法在重力坝材料参数反演中的应用[J].岩土力学,2007(10).
[23]程晓荣,张秋亮,王智慧等.基于人工鱼群算法的配电网网架优化规划[J].继电器,2007(21).
[24]袁远.基于改进人工鱼群算法的配电网无功优化[D].南京：南京理工大学,2008.
[25]李晓磊,薛云灿,路飞,田国会.基于人工鱼群算法的参数估计方法[J].山东大学学报(工学版),2004(3).
[26]Ma JianWei, Guo LiZhang. Applieation of artificialfish swarm neural net work in short term load fore easting of power system [J]. Power System Technology,2005(7).
[27]Djuric P M, Kotecha J H, Jianqui Zhang. Particle filtering[J]. Signal Processing Magazine,2003(5).
[28]姚玉霞.基于多智能体人工鱼群算法应用研究[D].西安：西安建筑科技大学,2008.
[29]陆陶荣,朱林户,李德芳.一种集群智能粒子滤波算法[J].西安电子科技大学学报,2008(3).
[30]莫以为,萧德云.进化粒了滤波算法及其应用[J].控制理论与应用,2005(2).
[31]李碧,林土胜.协同进化在遗传算法中的应用述评[J].计算机科学,2009(4).
[32]工凌,刘波.微粒群优化与调度算法[M].北京：清华大学出版社,2008.
[33]曾建潮,崔志华.微粒群算法[M].北京：科学出版社,2004.
[34]熊敏,刘玉树.基于协同进化遗传算法的地域选取方法[J].计算机技术与发展,2006(6).
[35]王书宇,李龙澍,汪群山.改进的协同进化遗传算法在机器博弈中的应用[J].计算机技术与发展,2008(12).
[36]李碧,林土胜.协同进化在遗传算法中的应用述评[J].计算机科学,2009(4).
[37]张创业,莫愿斌,何登旭.二群协同的人工鱼群优化算法[J].计算机技术与发展,2009(11).
[2]李晓磊,邵之江,钱积新.一种基于动物自治体的寻优模式：鱼群算法[J].系统工程理论与实践,2002(11).
[3]工闯.人工鱼群算法的分析及改进[D].大连：大连海事大学,2008.
[4]李晓磊,钱积新.基于分解协调的人工鱼群优化算法研究[J].电路与系统学报,2003(1).
[5]窦全胜,周春光,徐中宇等.群体启发进化规划[J].计算机科学,2005(7).
[6]栾丹.改进的进化规划算法[J].渤海大学学报(自然科学版),2005(2).
[7]Kramer K A, Stubberud S C. Tracking of multiple target types with a single neural extended kalman filter[J]. Intelligent Systems,2006(1).
[8]Khairnar Dilip Gopichand, Merchant S N. Nonlinear target identification and tracking using UKF[J]. Granular Computing,2007(1).
[9]袁泽剑,郑南宁,贾新春.高斯-厄米特粒子滤波器[J].电子学报,2003(7).
[10]Gordan N, Salmond D, Smith A. Novel approach to nonlinear/non-Gaussian Bayesian state estimation[J]. Proceedings on Radar and Signal Processing,1993(1).
[11]Amaud D, Simon G, Chistophe A. On sequential Monte Carlo sampling methods for Bayesian filtering[J]. Statistics and Computing,2000(10).
[12]Freitas J F G, Niranjan M, Gee A H. Sequential Monte Carlo methods to train neural network models [J]. Neural Computation,2000(4).
[13]K Murphy, S Russell. Rao-Blackwellised particle filtering for dynamic Bayesian networks[C]. New York:sequential Monte Carlo methods in practice,2001.
[14]Cheng C, Rashid A, Ashfaq K. Multiple object tracking with kernel particle filter[C]. 2005(1).
[15]Yang ShuYing, Ge WeiMin, Zhang Cheng. Tracking unknown moving targets on omnidirectional vision[J]. Vision Research,2009(3).
[16]徐宗本,张讲社,郑亚林.计算智能中的仿生学：理论与算法[M].北京：科学出版社,2003：9-20.
[17]Xiaojuan Shan, Mingyan Jiang, Jingpeng Li. The routing optimization based on improved artificial fish swarm algorithm[C]. Intelligent Control and Automation, 2006 (1).
[18]李亮,迟世春,林皋.禁忌鱼群算法及其在边坡稳定分析中的应用[J].工程力学,2006(3).
[19]张梅凤.人工鱼群智能优化算法的改进及应用研究[D].大连：大连理工大学,2007.
[20]黄光球,王西邓,刘冠.基于网格划分策略的改进人工鱼群算法[J].微电了学与计算机, 2007(7).
[21]高德芳,赵勇,郭杨.基于混合鱼群一蚁群算法的模块化产品配置设计[J].设计与研究,2007(1).
[22]宋志字,李俊杰,汪红宇.混沌人工鱼群算法在重力坝材料参数反演中的应用[J].岩土力学,2007(10).
[23]程晓荣,张秋亮,王智慧等.基于人工鱼群算法的配电网网架优化规划[J].继电器,2007(21).
[24]袁远.基于改进人工鱼群算法的配电网无功优化[D].南京：南京理工大学,2008.
[25]李晓磊,薛云灿,路飞,田国会.基于人工鱼群算法的参数估计方法[J].山东大学学报(工学版),2004(3).
[26]Ma JianWei, Guo LiZhang. Applieation of artificialfish swarm neural net work in short term load fore easting of power system [J]. Power System Technology,2005(7).
[27]Djuric P M, Kotecha J H, Jianqui Zhang. Particle filtering[J]. Signal Processing Magazine,2003(5).
[28]姚玉霞.基于多智能体人工鱼群算法应用研究[D].西安：西安建筑科技大学,2008.
[29]陆陶荣,朱林户,李德芳.一种集群智能粒子滤波算法[J].西安电子科技大学学报,2008(3).
[30]莫以为,萧德云.进化粒了滤波算法及其应用[J].控制理论与应用,2005(2).
[31]李碧,林土胜.协同进化在遗传算法中的应用述评[J].计算机科学,2009(4).
[32]工凌,刘波.微粒群优化与调度算法[M].北京：清华大学出版社,2008.
[33]曾建潮,崔志华.微粒群算法[M].北京：科学出版社,2004.
[34]熊敏,刘玉树.基于协同进化遗传算法的地域选取方法[J].计算机技术与发展,2006(6).
[35]王书宇,李龙澍,汪群山.改进的协同进化遗传算法在机器博弈中的应用[J].计算机技术与发展,2008(12).
[36]李碧,林土胜.协同进化在遗传算法中的应用述评[J].计算机科学,2009(4).
[37]张创业,莫愿斌,何登旭.二群协同的人工鱼群优化算法[J].计算机技术与发展,2009(11).