People counting with block histogram features and network flow constraints
Liqing Gao, Yanzhang Wang, Jian Wang · 2016
Recently, a directed graph model was presented to crowd counting in videos, and people flow was viewed as an integer flow on the constructed graph. The authors show that the network flow constraints on the graphs help to obtain consistent counting results. In this paper, we improve their work from two aspects. First, we design block histogram features for each group of people. Second, we simplify their directed graphs to contracted graphs by contracting all cut arcs. Since the network flow constraints on original graphs are equivalence to that on contracted ones, we propose a quadratic programming method with network flow constraints on contracted graphs to refine the crowd counting results. At last, experiments show that our block histogram features can deal better with perspective distortion problems and our approach obtains outstanding performance on PETS 2009 dataset.