关键词: 人群计数, 多尺度, 可信融合, 人群密度估计, 不确定性

Abstract: To solve the problem of large variation caused by different distances between monitoring camera and crowd in the crowd analysis tasks, a crowd counting framework based on trusted multi-scale fusion is proposed. Firstly, the framework extracts the common features with the traditional skeleton and then obtains the pedestrian information of different scales with the multi-scale information extraction module; Secondly, each scale network contains a population density estimation module and an uncertainty estimation module for evaluating the reliability of the prediction results of each scale; Then, the multi-scale prediction fusion module dynamically fuses the multi-scale prediction results based on the reliability to obtain more accurate density regression results. The experimental results show that the error of crowd counting is significantly reduced after the expansion of the existing method by multi-scale trusted fusion, which verifies the necessity, extensibility, rationality and effectiveness of the fusion strategy. In addition, different from the existing methods, the framework can not only predict the population density, but also provide the reliability of the prediction during the deployment stage, so that the algorithm can timely warn the inaccurate areas predicted by the model in practical application, and reduce the risk of avoiding the wrong prediction in subsequent analysis tasks.

Key words: crowd counting, multi-scale, trustworthy fusion, crowd density estimation, uncertainty