基于改进K-means聚类的惯性行人导航零速检测算法

Zero-Velocity Detection Algorithm for Inertial Pedestrian Navigation Based on Improved K-Means Clustering

行人导航中不同运动状态下零速区间的运动数据也有所不同,这就要求零速检测算法具有良好的适应性。针对利用阈值实现零速检测的算法在多种运动状态下适应性差的问题,该文提出了一种基于改进K-means聚类的零速检测算法(zero-velocity interval detection algorithm based on improved K-means clustering, IKC)。首先,在运动的开始阶段,通过K-means聚类对角速度数据进行聚类,从而得到零速区间与非零速区间的中心点;然后根据设定的数据点到中心点的距离条件对零速区间与非零速区间进行划分,相比于其他算法,优化了数据处理过程,有效缩短了计算时间,并且不依赖阈值条件,有效提高了该算法的适应性;同时,根据零速区间与非零速区间的持续时间判断运动状态是否改变,若发生改变则重新进行K-means聚类获取新运动状态的中心点。最后,在实际行人导航系统中对新提出的算法进行了实验验证,从计算量及行人导航精度等方面与步态特征提取的K均值聚类自适应判别算法(K-means clustering adaptive detection, KCA)、基于贝叶斯的自适应阈值零速检测算法(bayesian adaptive threshold detection, BAT)进行了对比分析。结果表明,本文提出的基于改进K-means聚类的零速检测算法不仅有效的减小了计算时间,而且具有较高的导航精度和导航稳定性。

In pedestrian navigation, the motion data of zero-velocity interval are different under different motion states, which requires the zero-velocity interval detection algorithm to have good adaptability. To solve the problem of poor adaptability of zero-velocity detection algorithm based on threshold in various motion states, this paper proposes a zero-velocity interval detection algorithm based on the improved K-means clustering(IKC). Firstly, at the beginning of the motion, the angular velocity data were clustered through K-means clustering, so as to obtain the center points of the zero velocity interval and the non-zero velocity interval. Then, the zero velocity interval and non-zero-velocity interval are divided according to the distance condition from the set data point to the center point. Compared with other algorithms, the data processing process is optimized, the calculation time is shortened effectively, and the algorithm is independent of the threshold condition, which effectively improves the adaptability of the algorithm. At the same time, according to the duration of zero-velocity interval and non-zero-velocity interval, whether the motion state has changed is judged. If there is a change, K-means clustering is performed again to obtain the center point of the new motion state. Finally, the proposed algorithm is validated in an actual pedestrian navigation system. In terms of computation amount and pedestrian navigation accuracy, this paper makes a comparative analysis with the K-means clustering adaptive discriminant algorithm for gait feature extraction(KCA)and the adaptive threshold zero-velocity interval detection algorithm based on Bayes(BAT). The results show that the zero-velocity interval detection algorithm based on the improved K-means clustering proposed in this paper not only effectively reduces the computation time, but also has higher navigation accuracy and stability.