基于LBP和HOG特征分层融合的步态识别

Gait recognition based on layered fusion of LBP and HOG features

针对步态能量图(GEI)在提取人体信息时只描绘出了轮廓信息,而丢失了内部特征的局限性,提出一种基于人体目标图像的局部二值模式(LBP)与方向梯度直方图(HOG)分层融合的GEI识别算法。该算法步骤包括:首先用光流法提取步态周期,获得一个周期的步态能量图(GEI);然后分三层提取GEI的LBP特征,得到三层的LBP图像;依次提取每层LBP图像的HOG特征,最后将每层提取的LBP和HOG特征融合,得到每层的新特征;最后将三个新特征依次融合成可以用于识别的最终特征。通过几种方法对CASIA和USF步态数据库的实验对比,提出的算法取得了更高的识别率。

For the Gait Energy Image （GEI）, which only depicts the contour information while extracting the human body information, and loses the limitation of the internal features, a GEI recognition algorithm based on the Local Binary Pattern （LBP） and the Histograms of Oriented Gradients （HOG） is proposed. The algorithm steps include：firstly, the gait cycle is extracted by using the optical flow method, and a cycle of the Gait Energy Image （GEI） is obtained, then the LBP features of GEI are extracted in three layers, the LBP images of three layers are extracted, then the HOG features of each layer are abstracted, and then the LBP and HOG features are fused. The new features of each layer are obtained, and finally three new features are fused into the final features that can be used for recognition. Through experimental comparison of CASIA and USF gait database, this method has achieved higher recognition rate.