A dense crowd tracking method using density map and optical flow

Traffic statistics in crowded places can provide support for passenger flow diversion. However, dense crowd statistics has problems such as occlusion, small scale, and complex background. Therefore, in this study, a method is proposed to use density maps for head target detection and optical flow for target tracking. In order to reduce human missed detection or false detection, this study designed two tracking algorithms based on trajectory prediction and target feature modeling. The experiment compared the tracking performance of two methods in scenarios with overlapping crowds and the effectiveness of re-tracking after target loss. The results show that feature-based tracking method has better tracking results and can be applied to estimate the number of people in dense crowds.