Abstract:With the rapid economic development and high concentration of urban population, people's income level and quality of life continue to improve, resulting in more and more crowded scenes caused by people going out. Especially in urban commercial centers, transportation hubs, sports venues during important events, tourist attractions, etc., crowd gatherings occur frequently. However, accidents involving crowd gatherings in public places occur frequently, causing heavy casualties and property losses. Therefore, for crowd recognition, this paper proposes a new method to accurately estimate the number of dense crowds. In this method, a density map with accurate pedestrian locations is first generated using the focal inverse distance transform and used as ground truth labels for network training. Then, a multi-scale feature fusion algorithm based on residual network is designed, combining spatial and channel attention mechanisms to improve the accuracy and stability of crowd density estimation. In dense crowds, the phenomenon of overlapping and occlusion of people is very common and serious, making it difficult for existing pedestrian detection methods to distinguish each individual and accurately count the flow of people. To solve this problem, this paper proposes a density map-based method that uses a local maximum detection strategy and a K-nearest neighbor algorithm to convert the density map into the corresponding dense head bounding box. This method can effectively reduce the impact of occlusion and improve the accuracy of people counting. In order to further improve the estimation accuracy, a pattern recognition density peak clustering algorithm is introduced to study the clustered crowds. By treating the head bounding box as an element point, the distance between each element point is calculated, and the density of each point is calculated. Then perform clustering to find the cluster center with the highest density in each class. Finally, by comparing the density of each cluster center with the corresponding density threshold and adopting the corresponding decision-making method, the accuracy of people counting is further improved.

Two-branch fusion network with attention map for crowd counting

Multi-branch Progressive Embedding Network for Crowd Counting

Dual Path Multi-Scale Fusion Networks with Attention for Crowd Counting

Dual-branch counting method for dense crowd based on self-attention mechanism

LEVERAGE MULTI-SCALE DILATED CONVOLUTIONAL NEURAL NETWORK WITH GLOBAL ATTENTION FEATURE FUSION FOR CROWD COUNTING

Multi-scale features fused network with multi-level supervised path for crowd counting

Crowd density estimation based on multi scale features fusion network with reverse attention mechanism

Double multi-scale feature fusion network for crowd counting

MLANet: multi-level attention network with multi-scale feature fusion for crowd counting

A Crowd Counting and Localization Network Based on Adaptive Feature Fusion and Multi-Scale Global Attention Up Sampling

FDCNet: Frontend-Backend Fusion Dilated Network Through Channel-Attention Mechanism

Multi-feature Counting of Dense Crowd Image Based on Multi-column Convolutional Neural Network

Crowd Density Estimation Using Fusion of Multi-Layer Features

Multi-scale fusion public gathering recognition based on residual network

A multi-scale fusion and dual attention network for crowd counting

BBA-NET: A Bi-Branch Attention Network for Crowd Counting

MGSNet: A Multi-Scale and Gated Spatial Attention Network for Crowd Counting

Single-column CNN for crowd counting with pixel-wise attention mechanism

Cascade-guided multi-scale attention network for crowd counting

Counting with Self-Weighted Multi-Scale Fusion Networks

A High-Density Crowd Counting Method Based On Convolutional Feature Fusion