陈建华, 奚如如, 王兴松, 杜峰坡, 穆希辉. 外骨骼机器人的非结构地面行走步态分类算法[J]. 机器人, 2017, 39(4): 505-513. DOI: 10.13973/j.cnki.robot.2017.0505
引用本文: 陈建华, 奚如如, 王兴松, 杜峰坡, 穆希辉. 外骨骼机器人的非结构地面行走步态分类算法[J]. 机器人, 2017, 39(4): 505-513. DOI: 10.13973/j.cnki.robot.2017.0505
CHEN Jianhua, XI Ruru, WANG Xingsong, DU Fengpo, MU Xihui. Walking Gait Classification Algorithm for Exoskeleton Robot on Unstructured Ground[J]. ROBOT, 2017, 39(4): 505-513. DOI: 10.13973/j.cnki.robot.2017.0505
Citation: CHEN Jianhua, XI Ruru, WANG Xingsong, DU Fengpo, MU Xihui. Walking Gait Classification Algorithm for Exoskeleton Robot on Unstructured Ground[J]. ROBOT, 2017, 39(4): 505-513. DOI: 10.13973/j.cnki.robot.2017.0505

外骨骼机器人的非结构地面行走步态分类算法

Walking Gait Classification Algorithm for Exoskeleton Robot on Unstructured Ground

  • 摘要: 针对人体运动随机性和非结构地面等因素造成在不同类型地面上足底力(GRF)差异大的情况,研发了一套配有鞋底压力传感器、用于实时检测足底力变化的实验靴,提出了基于PSO-SVM(基于粒子群优化算法的支持向量机)的步态分类方法.根据足底受力云图,该实验靴中冗余布置了7枚压力传感器.对人行走在步行机(5 km/h)、水平硬路面和野外草地上的足底力进行了采集和处理.将基本组的足底力作为训练集,预设对应的标签值.基于这些训练集,构建了普通的分类器I和基于粒子群优化算法的支持向量分类器II,并分别利用它们对水平硬路面和野外草地行走的数据进行分类检验.实验结果表明,与普通的分类器I相比,该优化算法不仅对复杂地面上的行走步态分类具有明显的优势,对水平硬路面和野外草地的行走步态识别正确率分别提升了32.9%和42.8%,而且能在某些鞋底传感器发生故障后保持较快的寻优速度和较好的鲁棒性.

     

    Abstract: For ground reaction force (GRF), there exists huge difference on different terrains because of human leg's random movement and unstructured ground. Therefore, an experiment shoe with insole-type pressure sensors is developed for detecting GRF in real time, and a PSO-SVM (particle swarm optimization based support vector machine) algorithm for gait classification is presented. 7 pressure sensors are installed into the shoes redundantly according to the plantar pressure distribution. The GRFs when a person walks on treadmill (5 km/h), concrete pavement and field grass are collected and processed. Corresponding labels are set in advance for the GRFs from the basic group, which is used as training set. Based on the training set, an ordinary classifier I and a classifier Ⅱ based on PSO-SVM algorithm are constructed to classifying the walking data on concrete pavement and field grass respectively. The experiment results demonstrate that, compared with the ordinary classifier I, the proposed optimization algorithm shows a great advantage of gait classification on complex terrains, gait identification accuracies on concrete pavement and field grass are improved by 32.9% and 42.8% respectively, and still a fast optimization speed and a good robustness are implemented after some insole-type pressure sensors malfunction.

     

/

返回文章
返回