根据移动机器人系统多目标跟踪的特点,提出了一种包括运动检测、数据关联和目标状态估计3个层次的多目标跟踪混合条件随机场.作为一种辨别式模型,该混合条件随机场模型允许状态与数据之间存在非局部的依赖关系,不仅可以利用目标的运动信息和局部形状信息提高多目标跟踪中数据关联的精度,而且可以利用多次观测数据检测新目标,实现新目标检测与目标跟踪的同步.在自行研发的移动机器人平台上的多目标跟踪实验结果表明,基于文中提出的混合条件随机场的移动机器人多目标跟踪方法比基于产生式模型的方法JPDAF具有更高的精度与稳定性.

According to the characteristics of multi-object tracking of mobile robots,a hybrid conditional random field（HCRF） model,which consists of three layers including the motion detection layer,the data association layer and the state estimation layer,is proposed.As a kind of discriminative model that allows nonlocal dependencies between the state and the observation data,the proposed model can not only utilize local motion information and shape information to improve the accuracy of data association according to the motion information and the local shape information of the object,but also integrate object tracking and moving object detection according to the observation data obtained in multiple time steps.Experimental results of the multi-object tracking of the self-developed mobile robot show that the tracking based on the proposed HCRF model is more accurate and stable than that of the JPDAF method based on the generative model.