基于交互式多模型Kalman滤波的手势跟踪算法

Gesture Tracking Algorithm Using Interactive Multi-model Kalman Filter

针对任意变形手势跟踪过程中,手势运动轨迹方向发生改变时,传统滤波跟踪算法跟踪精度迅速下降的问题,提出了一种基于交互式多模型Kalman滤波的改进型手势跟踪算法。该算法在传统非机动状态空间模型的基础上增添了两个机动模型,以更加准确的描述手势的状态空间。首先采用三个不同模型分别描述不同的目标运行模式;同时,以模型匹配似然函数为基础更新模型概率;最终组合所有滤波器修正后的状态估计值以得到最优状态估计。实验对比结果表明,该算法能够取得较高地跟踪精度。

In the process of gesture tracking for arbitrary deformation gesture,the tracking precision of traditional tracking filter algorithm drops rapidly when the gesture trajectory direction is changed.To solve this problem,an improved gesture tracking algorithm based on interactive multi-model Kalman filtering is proposed,which can describe the gesture s state space more accurately by adding two maneuvered models to the traditional non-maneuvered state space model.Firstly,three different models are used to describe different target operating modes respectively.At the same time,the model probability is updated based on the model matching likelihood function.Finally,all the filter-corrected state estimates are combined to obtain the optimal state estimate.Experimental results show that the algorithm can achieve higher tracking accuracy.