基于自适应特征卷积网络的行人检测方法

Pedestrian detection method based on adaptive feature convolution network

针对行人检测方法未能充分利用卷积网络浅层特征的问题,改进Faster R-CNN框架,提出了一种基于自适应特征卷积网络的行人检测方法。该方法有两处改进:a)设计了SFCM模块,用于提取卷积神经网络浅层细节特征;b)引用挤压与激励操作设计了AFCM模块,用于筛选检测所需的强辨识力行人特征。此外,利用公开的Caltech和INRIA行人数据集,通过在基准框架中逐一添加SFCM和AFCM模块训练行人检测器,验证了所提模块的有效性,并对比了主流行人检测算法。实验结果显示,所提方法的误检率分别降到了9.13%和9.46%,具有更优的检测性能。

To circumvent the problem of failing to make full use of the shallow features of the convolutional network,this paper improved the existing Faster R-CNN framework and proposed pedestrian detection method based on adaptive feature convolution network,for achieving higher the detection accuracy.This paper has two improvements.Firstly,it designed SFCM module to extract the shallow detail features of the convolution neural network.Secondly,it proposed AFCM module by utilizing the squeeze and excitation mechanism,which was used to screen the strong discrimination features of pedestrian.Moreover,it used two public pedestrian datasets,Caltech and INRIA.It added SFCM module and AFCM module one by one in the benchmark framework,which verified the validity of the designed modules.Compared with some existing person detection algorithms,the experimental results show that the proposed method has better detection performance and miss rate dropped to 9.13%and 9.46%respectively.