Dense Crowd Counting Based on ResNet

Abstract

High-density crowd gathering is very prone to various accidents, so real-time monitoring and analysis of dense crowds to prevent accidents is of great practical significance. In this paper, the density crowd detection counting is implemented based on the fine-tuned optimization of the ResNet model, and the evaluation and warning function is added. The average absolute error of the comprehensive performance index obtained after the final training model test reaches 7.9, that is, each prediction result is controlled within $\pm {\mathrm {7.9}}$ of the correct value, which proves that the model can effectively count high-density crowds and give evaluation and warning results.