多尺度空间金字塔池化PCANet的行人检测

Pedestrian Detection Using Multi-scale Principal Component Analysis Network of Spatial Pyramid Pooling

针对非理想条件下行人检测的性能和效率问题,提出多尺度空间金字塔PCANet。将空间金字塔作为网络的特征池化层,通过分层池化特征的方式获得图像的显著性特征,并将底层特征和高层特征级联以获得样本的多尺度特征的向量表示,输入SVM分类器。在INRIA和NICTA数据库中,与HOG、CNN等算法进行行人检测对比实验,结果表明,该算法有更高的正确检测率、更低的漏检率和误检率。

Pedestrian detection is easily affected by non-ideal factors such as complex background,shooting angle and diversity of human body posture in natural environment.To solve this problem,this paper proposes a Multi-scale Principal Component Analysis Network of Spatial Pyramid Pooling (MS-PCANet-SPP).The feature pooling layer using spatial pyramid pooling method extract the saliency features of the image.The multi-scale features of the input samples can be obtained by cascading the high-level and low-level features,which is input to the SVM classifer.The comparative experiments are performed in the INRIA and NICTA databases.Experimental results show that,compared with HOG,CNN and other algorithms,MS-PCANet-SPP has a higher detection rate,a lower miss rate,and a lower false positive rate.