Crowd Counting via Scale-Adaptive Convolutional Neural Network

The task of crowd counting is to automatically estimate the pedestrian number in crowd images. To cope with the scale and perspective changes that commonly exist in crowd images, state-of-the-art approaches employ multi-column CNN architectures to regress density maps of crowd images. Multiple colum...

Full description

Saved in:
Bibliographic Details
Published in2018 IEEE Winter Conference on Applications of Computer Vision (WACV) pp. 1113 - 1121
Main Authors Zhang, Lu, Shi, Miaojing, Chen, Qiaobo
Format Conference Proceeding
LanguageEnglish
Published IEEE 01.03.2018
Subjects
Computer architecture
Feature extraction
Head
Spatial resolution
Switches
Task analysis
Training
Online AccessGet full text

Cover

More Information
Summary:The task of crowd counting is to automatically estimate the pedestrian number in crowd images. To cope with the scale and perspective changes that commonly exist in crowd images, state-of-the-art approaches employ multi-column CNN architectures to regress density maps of crowd images. Multiple columns have different receptive fields corresponding to pedestrians (heads) of different scales. We instead propose a scale-adaptive CNN (SaCNN) architecture with a backbone of fixed small receptive fields. We extract feature maps from multiple layers and adapt them to have the same output size; we combine them to produce the final density map. The number of people is computed by integrating the density map. We also introduce a relative count loss along with the density map loss to improve the network generalization on crowd scenes with few pedestrians, where most representative approaches perform poorly on. We conduct extensive experiments on the ShanghaiTech, UCF CC 50 and WorldExpo'10 datasets as well as a new dataset SmartCity that we collect for crowd scenes with few people. The results demonstrate significant improvements of SaCNN over the state-of-the-art.
DOI:10.1109/WACV.2018.00127