DEO-Net: Joint Density Estimation and Object Detection for Crowd Counting

Automated crowd counting has emerged as a vision-based measurement method for crowd analysis and management. However, current methods based on density maps still suffer from challenges related to background noise and blurring effects. To address the limitations, this work proposes a deep neural network, named joint density estimation and object detection (DEO-Net), specifically designed to generate high-quality density estimation maps. DEO-Net bridges the gap between detection and density estimation-based methods in crowd counting. The key contributions of this research are as follows: 1) DEO-Net incorporates object detection for more accurate crowd localization; 2) the network training is optimized with an independent structural similarity index (I-SSIM) and curriculum losses to better learn local structural information and recognize local maxima; and 3) the experimental results demonstrate the state-of-the-art (SOTA) performance of the proposed DEO-Net with mean absolute error (MAE) values of 54.2, 6.2, 83.1, and 57.3 on the ShangHaiTechA, ShanghaiTechB, UCF_QNRF, and JHU-CROWD++ public datasets, respectively.