模型关系有向图辅助下的KL-VSIMM目标跟踪算法

KL-VSIMM TARGET TRACKING ALGORITHM ASSISTED BY MODEL RELATION DIRECTED GRAPH

为提高交互多模(Interacting Multiple Model,IMM)算法模型选择的灵活性,减少模型间的竞争,提出一种变结构交互多模(Variable-Structure Interacting Multiple Model,VSIMM)算法。在IMM算法的基础上计算当前各模型的KL值,并作为参考量评估模型与目标当前运动模式的接近程度,计算出与目标运动模式相似度较高的模型;借助模型关系有向图激活其在加速度空间上的邻近模型,同时剔除匹配程度最低的模型以实现模型集的自适应切换,形成KL-VSIMM算法。仿真结果表明,KL-VSIMM算法在状态估计质量和模型识别能力上均表现出了很好的性能。

In order to improve the flexibility of IMM algorithm model selection and reduce the competition among models,we propose a new variable-structure interacting multiple model(VSIMM)algorithm.Based on the IMM algorithm,we calculated the current(Kullback-Leiber,KL)values of each model,and used it as a reference to evaluate the proximity of the model to the current motion pattern of the target.The model with high similarity to the target motion pattern was calculated,and its adjacent model in acceleration space was activated with the help of the model relational directed graph.The model with the lowest matching degree was eliminated to realize the adaptive switching of model set,and the KL-VSIMM algorithm was formed.The simulation results show that the KL-VSIMM algorithm has good performance in both the state estimation quality and the model recognition ability.