基于双鉴别器策略的空间感知型人群计数算法

Spatial-aware crowd counting algorithm based on dual discriminator strategy

人群计数是计算机视觉中的重要研究领域,为了精确计算静态图像或视频帧中的行人数,通常采用回归密度图法用于提高计数精度,然而,由于源域和目标域之间的域移位问题,模型在稀缺标签数据集上的泛化能力不足。文章提出了一种基于双鉴别器策略的空间感知型(spatial awareness of dual discriminator strategy,SADDS)域适应人群计数算法,旨在有效解决图像风格和人头尺度的差异。该算法在DSANET骨干网络VGG16的特征提取过程中融合了分层注意力机制(hierarchicalattentionmechanism,HAM),包括空间注意力机制(spatial attention mechanism,SAM)和全局注意力机制(global attention mechanism,GAM),同时采用全局和局部双鉴别器来减少域移位。通过在公开数据集上的实验,验证了该方法的有效性与优越性。

Crowd counting is an important research field in computer vision.In order to accurately calculate the number of rows in static images or video frames,the regression density map method is usually used to improve the counting accuracy.However,due to the domain shift problem between the source domain and the target domain,the generalization ability of the model on the scarce label data set is insufficient.This paper proposes a space-aware spatial awareness of dual discriminator strategy(SADDS)domain adaptive crowd counting algorithm based on dual discriminator strategy,which aims to effectively solve the difference between image style and head scale.The algorithm combines hierarchical attention mechanism(HAM),including spatial attention mechanism(SAM)and global attention mechanism(GAM),in the feature extraction process of DSANET backbone network VGG16.At the same time,global and local double discriminators are used to reduce domain shift.The effectiveness and superiority of the method are verified by experiments on public data sets.