基于MPSO-BP神经网络方法的人体步态识别

为提高人体下肢步态相位识别准确率以实现外骨骼机器人控制，采用一种改进的粒子群优化MPSO-BP神经网络方法识别不同运动模式下的人体步态相位。通过自适应调整学习因子构造MPSO-BP神经网络分类器，以多种传感信息组成的特征向量样本集训练神经网络分类器，用于识别人体下肢在平地行走、上楼梯和起坐三种典型运动模式下的步态相位。实验结果表明，MPSO-BP神经网络分类器能有效识别三种不同运动模式的步态相位，识别准确率均达到96%以上，识别性能优于传统的BP神经网络模型和粒子群优化神经网络模型。

Human gait recognition based on MPSO-BP neural network method

To improve the accuracy rate of human gait phase recognition for controlling the exoskeleton robot, an approach based on Modified Particle Swarm Optimization algorithm-Back Propagation（MPSO-BP） neural network is utilized to divide three types of gait into different phases. Firstly, the MPSO-BP neural network classifier is constructed through regulating the learning factor adaptively, and then the classifier is trained using sample set containing multi-sensor information. Secondly, test the classifier on gait phase recognition in three types of human gait including walk, upstairs and sit-down. The experimental results show that the MPSO-BP neural network classifier can successfully increase the accuracy rate up to averaged 96% above, which is superior to the BP neural network and the particle swarm optimization BP neural network methods.