基于步态能量图与VGG结合的步态识别方法

Gait Recognition Method Based on Gait Energy Map Combined with VGG

基于深度学习端到端和多层特征提取的思想,给出一种基于步态能量图和VGG卷积神经网络结合的步态识别方法。首先,使用背景减除法分割出人体轮廓;然后,通过身体轮廓宽度变化计算出步态周期;其次,根据步态周期图像计算出步态能量图;最后使用VGG网络对步态能量图进行特征学习及分类。实验结果表明:所提出的方法可以准确识别行人身份,在CASIA-B步态数据中平均准确率可达92.5%,且对视角有较好的鲁棒性,对深度学习在步态识别领域的进一步应用有借鉴意义。

Based on the idea of end-to-end deep learning and multi-layer feature extraction,this paper adopts a gait recognition method based on gait energy map and VGG convolutional neural network. Firstly,the background contour is used to segment the human body contour,and then the gait cycle is calculated by the body contour width variation. Secondly,the gait energy map is calculated according to the gait cycle image. Finally,VGG network is used to study and classify gait energy map. The experimental results show that the proposed method can accurately identify pedestrian identity,and the average accuracy rate in the CASIA-B gait data can reach 92. 5%,and it has better robustness to the viewing angle. It can be used for reference in the further application of deep learning in gait recognition.