Pay Attention To Deep Feature Fusion In Crowd Density Estimation

Huimin Guo,Fujin He,Xin Cheng,Xinghao Ding,Yue Huang
DOI: https://doi.org/10.1007/978-3-030-36808-1_39
2019-01-01
Abstract:Crowd density estimation has important practical significance for effectively suppressing the occurrence of stampede accidents. However, the crowd counting task can be easily interfered by various factors such as perspective, congestion, occlusion, density, etc., which makes accurate crowd counting a challenging task. To solve these problems, in this paper, we propose an effective hierarchical aggregation module to fuse different scale information in the network. Since the crowd counting task is seriously interfered by the surrounding environment, in this paper we propose to use attention mechanism module to weight the spatial position of the network learned feature map to effectively limit the interference of the background region to the crowd counting task. Finally, a large number of related experiments show that our model in this paper has strong generalization ability while having better performance on several public datasets compared to existing model algorithms.
What problem does this paper attempt to address?