基于改进的卷积神经网络的步态识别

为了提高惯性传感器采集到的序列数据中步态识别的准确率,建立了一个激励层改进的卷积神经网络(CNN)模型。针对三轴加速度传感器对运动太过敏感导致步态周期划分不准确的问题,采用加速度传感器与弯曲度传感器组合获取人体运动信息。将CNN模型中激励层的线性整流函数(ReLU)改进为带泄露线性整流函数(Leaky ReLU),以解决遇到卷积输出数据小于0时神经元被抑制的问题,进而达到提高步态识别准确率的目的。实验结果表明激励层优化的CNN模型在行走、上下楼和上下坡五种步态模式下识别率达到了95.79%,与未采用弯曲度传感器的改进CNN模型和未进行激励层改进的CNN模型相比,步态识别率有所提高。

Gait Recognition Based on Improved Convolutional Neural Network

In order to improve the accuracy of gait recognition in the sequence data collected by inertial sensors,an improved convolutional neural network(CNN)model with excitation layer is established.Aiming at the problem that the triaxial accelerometer is too sensitive to motion,resulting in inaccurate gait cycle division,the combination of triaxial accelerometer and bending sensor is used to obtain human motion information.The rectified linear unit(ReLU)of the excitation layer in CNN model is improved to leaky ReLU to solve the problem that neurons are suppressed when the convolution output data is less than zero,so as to improve the accuracy of gait recognition.The experimental results show that the recognition rate of the CNN model optimized by the excitation layer reaches 95.79%in five gait modes:walking,going upstairs and downstairs and going uphill and downhill.Compared with the improved CNN model without bending sensor and the CNN model without excitation layer improvement,the gait recognition rate is improved.