Statistical Calculation of Dense Crowd Flow Antiobscuring Method considering Video Continuity

People flow statistics have important research value in areas such as intelligent security. Accurately identifying the occluded target video surveillance is a difficulty system. Now popular moving object tracking algorithm based on detection and cannot accurately determine relationship between overlapping. For of people system, dense crowd antiocclusion statistical considering continuity proposed. This study focuses improved faster R-CN for small detection, correlation matching, two-way human statistics. According to small-scale characteristics head target, R-CNNV network structure adaptively improved. The shallow images features are used improve feature extraction ability targets. occlusion function constructed clearly express targets, it incorporated into framework algorithm. A trajectory prediction follow targets real time, method accomplish flow. To prove strength method, tests managed scenes with different degrees density, results show that improves average accuracy 7.31% 10.71% Brainwash test set Pets2009 benchmark data set, respectively, compared original F-value comprehensive evaluation index stream various can reach more than 90%. Compared excellent methods SSD sorting yolov3 deepsort recent years, its F increased by 1.14%–3.04%.