一种基于行人步态数据集的步长估计算法

A Stride Length Estimation Algorithm Based on Pedestrian Gait Data Set

步长估计是步行者航位推算的关键问题,但是惯性传感器的各类测量噪声会导致航位推算误差的累积。而步态分割可以在时空维度提供更细致、精确的分析,因此适用于定位领域的步长估计。首先建立了行人步态数据集,包括26位不同身高的行人个体使用3档不同速度行走的步态数据与步态分割的标签;然后基于该数据集,采用时间规整算法训练步伐检测模型,实现步态识别算法,步态识别F值达到0.701;最后在步态识别分析的基础上,采用梯度提升决策树模型,一定程度下实现了跨速域、跨个体的步长估计。

Stride length estimation is one of the key elements in pedestrian indoor positioning.However all kinds of measurement noise of inertial sensor will lead to the accumulation of dead reckoning error.Gait segmentation can provide more detailed and accurate analysis in the space-time dimension,so it is suitable for step estimation in the field of localization.Firstly,this paper establishes the stride gait data set,including the gait data and gait segmentation labels of 26 pedestrian individuals with different heights walking at three speeds.Then,based on the data set,the time warping algorithm is used to train the step detection model and realize the gait recognition.The F value of gait recognition reaches 0.701.Finally,based on the analysis of gait recognition,the gradient lifting decision tree model is used to realize the step estimation across speed domain and individuals to a certain extent.