第5卷第2期 智能系统学报 Vol.56.2 2010年4月 CAAI Transactions on Intelligent Systems Apr.2010 doi:10.3969/i.issn.1673-4785.2010.02.006 一种机器人未知环境下动态目标跟踪 交互多模滤波算法 伍明，孙继银 (中国人民解放军第二炮兵工程学院计算机应用系，陕西西安710025) 摘要：为了解决机器人同时定位、地图构建和目标跟踪问题，提出了一种基于交互多模滤波(interacting multiple model filter,IMM)的方法.该方法将机器人状态、目标状态和环境特征状态作为整体来构成系统状态向量并利用全 关联扩展式卡尔曼滤波算法对系统状态进行估计，由此随着迭代估计的进行，系统各对象状态之间将产生足够的相 关性，这种相关性能够正确反映各对象状态估计间的依赖关系，因此提高了目标跟踪的准确性.该方法进一步和传 统的IMM滤波算法相结合，从而解决了目标运动模式未知性问题，IMM方法的采用使系统在完成目标追踪的同时 还能对其运动模态进行估计，进而提高了该算法对于机动目标的跟踪能力.仿真实验验证了该方法对机器人和目标 的运动轨迹以及目标运动模态进行估计的准确性和有效性. 关键词：IMM滤波；EKF滤波；同时定位；地图构建；目标跟踪；移动机器人 中图分类号：TP242.6文献标识码：A文章编号：16734785(2010)020127-12 An interacting multiple model filtering algorithm for mobile robots to improve tracking of moving objects in unknown environments WU Ming,SUN Ji-yin (Department of Computer,The Second Artillery Engineering College,Xi'an 710025,China) Abstract:A novel method was developed for synchronous localization and mapping (SLAM)and object tracking (OT)to provide simultaneous estimation of a robots and any objects trajectories in an unknown environment.The system was based on interacting multiple model (IMM)filtering.In this approach,the states of robots,objects and landmarks were used to form an integrated system state.A full covariance extended Kalman filter (EKF)was then employed to estimate system state.As the iterative estimation progressed,sufficient correlations between the differ- ent objects in the system could be establish to reflect the interdependent relationships of estimations between differ- ent system objects.In this way the precision of object state estimation was improved.Moreover,when combined with a traditional IMM filter algorithm,this method solved the uncertainty problem for modes of object motion.With the application of IMM,the method helped robots to track objects and estimate their modes of motion,improving the accuracy of object localization.Simulation results validated the effectiveness of the proposed method in the esti- mation of the trajectories of robots and objects and the modes of motion of tracked targets. Keywords:interacting multiple model filter;extended Kalman filter;simultaneous localization and mapping;object tracking;mobile robot 机器人同时定位与地图构建(SLAM)是指机器 程.SLAM的难度在于准确的地图构建依靠于准确 人在未知环境下，根据传感器信息迭代地完成环境 的机器人位姿估计，而准确的机器人位姿估计反过 地图构建，并同时对机器人位姿状态进行估计的过 来又依靠于准确的环境地图.对于这个“鸡生蛋，蛋 生鸡”的问题主要存在2类解决方法，其一是基于 收稿日期：200908-30. 通信作者：伍明.E-mail:hyacinth531@163.com 扫描点匹配的方法13]，其二是基于Bayes估计的方