基于HOG特征的步态能量图身份识别算法

由于步态能量图像(GEI)是对二值轮廓图像序列相加求平均,然而,二值轮廓图像只能捕获人体轮廓的边界信息,人体的内部边界信息会被完全的丢弃掉,基于GEI算法的缺陷,提出了一种基于人体目标图像的方向梯度直方图(HOG)特征的GEI识别算法,此算法不仅能捕获人体轮廓的边界信息,而且还能提取人体重合的边界信息.获取人体目标图像的HOG特征的步态能量图,首先使用视频前景分割算法提取人体目标图像,然后提取图像序列中每帧人体目标图像的HOG特征;最后对图像序列中的每帧HOG特征图像相加求平均.在此基础上,依据GEI和HOG的思想,又实现了对传统步态能量图、二值轮廓图像序列、人体目标图像步态能量图进行HOG特征提取及直接构建人体目标图像步态能量图特征的表示,从而提出了 4种拓展的能量图构建方法,并针对这5种算法与经典的GEI算法利用CASIA步态数据库进行了实验分析对比,实验结果表明算法效果良好.

Base on gradient histogram energy image algorithm for person identification

In GEI,because binarized silhouettes are averaged over full gait sequence,binarized silhouettes can only capture edge information at the boundary of the person.Based on the defects of GEI algorithm,a gradient histogram energy image algorithm for person identification was proposed.the gradient histogram energy image (GHEI) can capture edge information at the boundary of the person,but also captures edges within the person by means of gradient histograms.The process of gradient histogram energy image on the foreground of each frame (FEF-GHEI) calculation can be detailed as follows:First,the foregrounds are segmented from each frame.And then,we calculate HOG on the foreground of each frame separately.Finally,the resulting gradient histograms are averaged over full gait cycles.On this basis,according to the idea of gait energy image (GEI) and histograms of oriented gradients (HOG),the four variations of energy images proposed.The proposed five experiments and gait energy image (GEI) were run on the widely used CASIA gait database and carry on the analysis comparison,The proposed methods show significant performance improvements over the current state of the art.