基于重构吸引子融合BAG方法的步态识别研究

The Research of Gait Recognition Based on Fusion of Reconstructed Attractors and BAG Method

针对基于惯性传感器的步态识别方法在动态情况下表现不佳且计算复杂度较高的问题,提出一种基于重构吸引子融合盒近似几何(BAG)方法。首先,将人类步态视作一个动态系统,根据Taken理论在潜在空间重构吸引子;然后,利用奇异谱分析方法获得奇异值,并将其应用于惯性传感器的标量测试;最后,利用盒近似几何方法完成步态识别。针对20个不同对象的模式分析了各种参数对步态识别性能的影响,实验结果表明,相比其它几种步态识别方法,本文方法能够实现高精度的识别且具有较低的计算复杂度。

For the problem of poor performance in ambulatory conditions and high computational complexity of current gait recognition methods based on inertial sensors, a novel gait recognition method is proposed. Firstly, the human gait is considered as a dynamical system and attractor is reconstructed in latent space according to the Taken theory. Then, the singular spectrum analysis is applied to scalar measurements from the inertial sensor. Finally, the gait recognition method based on reconstructed attractors in latent space and box approximation geometry is proposed. Effects of different parameters on the gait recognition performance using patterns from 20 different subjects are analyzed. The experiment results show that the method proposed has high accuracy and low computational complexity.