基于分层梯度方向直方图和SVM的人体识别

Human Recognition Based on Hierarchical Histograms of Oriented Gradients and Support Vector Machines

为了进一步提高图像人体识别精度,提出了一种能够反映局部图像整体形状信息的加权分层梯度方向直方图(HOG)特征;采用主成分分析(PCA)法对所提出的特征结构进一步改进,得到了另一种分层HOG+PCA特征;应用了基于径向基函数(RBF)核的支持向量机(SVM)模型作为人体分类器,交叉验证了模型的参数,并在一个较为完备的人体图像样本集上进行了训练和测试;实验结果表明,所得到的两种局部图像形状描述特征均具有比HOG特征更高的人体识别精度。

In order to further improve the precision of human recognition,this paper presents a new feature called weighted hierarchical histograms of oriented gradients（HOG）,which can reflect the overall shape information of the local image.Principal component analysis（PCA） algorithm is applied to improve the structure of the proposed feature,and get another new feature called hierarchical HOG＋PCA;Support vector machines（SVM） based on radial basis function（RBF） kernel is selected as human classifier,and its best parameters are got by the cross-validation procedure.The experiments conducted on the complete training and test sample set show that both of the recognition precisions of the proposed features outperform the ones of the HOG.