A Survey on Crowd Counting Methods and Datasets

摘要

Recent successful works of crowding counting are introduced. We summarize several classic achievements of traditional methods: regression methods, detection methods, and density map estimation methods. Some CNN models are categorized according to its function and structure. Specially, we discuss some problems have solved by CNN models like different scale, different background, and lack of label. CNN methods rely highly on the dataset, so several classic and popular datasets and some newly released dataset are presented. At last, we recognized probably the most convincing difficulties and issues which are investigated in crowd counting and density estimation utilizing computer vision and machine learning methods.