基于卷积神经网络的步态周期检测方法  

Gait period detection method based on convolutional neural networks

作　　者：王科俊 刘亮亮[1] 丁欣楠 胡钢 徐怡博 WANG Kejun;LIU Liangliang;DING Xinnan;HU Gang;XU Yibo(College of Intelligent Systems Science and Engineering, Harbin Engineering University, Harbin 150001, China)

机构地区：[1]哈尔滨工程大学智能科学与工程学院,黑龙江哈尔滨150001

出　　处：《哈尔滨工程大学学报》2021年第5期656-663,共8页Journal of Harbin Engineering University

基　　金：国家自然科学基金项目(61573114).

摘　　要：针对步态周期检测直接影响到步态识别的计算量和精度的问题,本文基于深卷积神经网络实现步态周期检测,分别通过将步态序列根据周期性进行分类,和将步态序列拟合为正弦函数进行步态周期性检测的方法,对步态周期进行建模。其关键思想是根据步态周期的规律性,将步态起伏作为分类问题或一个正弦函数来建模,步态视频中的每个帧对应一个可以表示其周期特征的类别或函数值。而卷积神经网络被用于提取步态帧的周期性特征,以定位该帧在周期中的位置,得到分类或回归结果,最终实现步态周期检测。在CASIA-B数据集的不同视角下采用了多种网络结构进行了周期检测效果的验证。实验结果表明:该方法对步态周期性检测具有良好的精度和鲁棒性。Gait period detection directly affects the computational cost and accuracy of gait recognition.On the basis of deep convolutional neural networks,this paper models the gait cycle by classifying the gait sequence according to periodicity and fitting the gait sequence to sine functions to detect the gait cycle.The key is to abstract gait period detection as a classification problem or a sine function based on the regularity of the gait period.Each frame in a gait video corresponds to a category or a function value that can represent its periodic characteristics.Convolutional neural networks are used to extract periodic features of the gait frame,locate the position of the frame in the cycle,obtain the classification or return result,and finally realize gait cycle detection.In the CASIA-B dataset,we test a variety of network structures in terms of different views to verify the effect of periodic detection.The experimental results are compared with those of other state-of-the-art works and show that the proposed method achieves good accuracy and robustness for gait period detection.

关 键 词：步态周期检测 步态序列 卷积神经网络 深卷积神经网络 步态识别 生物特征识别 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]