Pedestrian Counting via Deep Convolutional Neural Networks In Crowded Scene

DOI

10.2991/amitp-16.2016.67How to use a DOI?

Keywords

pedestrian, counting, crowd, CNN, EIN.

Abstract

Currently pedestrian counts mainly faces two major problems in the crowd scene: expression of pedestrian's features and perspective shade. To address this problem, we propose a deep convolutional neural network (CNN) and Ensemble Inference Network (EIN) for crowd pedestrian recognizing and counting. First, we propose a Perspective-Correct Interpolation model for extract more features in crowded scenes. Then, we design a convolution layer which contains preventing occlusion layer in the neural network. Finally to achieve lower error rates, our CNN-based method introduces Ensemble Inference Network to the training and classification processes. Experimental results show that the error rate of the proposed method is 0.28 and possess accurate recognition in pedestrian comparable performance to state-of-the-art methods in crowded scenes.

Copyright

© 2016, the Authors. Published by Atlantis Press.

Open Access

This is an open access article distributed under the CC BY-NC license (http://creativecommons.org/licenses/by-nc/4.0/).

Download article (PDF)

TY  - CONF
AU  - Jingwei Li
AU  - Jianxin Song
PY  - 2016/09
DA  - 2016/09
TI  - Pedestrian Counting via Deep Convolutional Neural Networks In Crowded Scene
BT  - Proceedings of the 2016 4th International Conference on Advanced Materials and Information Technology Processing (AMITP 2016)
PB  - Atlantis Press
SP  - 341
EP  - 347
SN  - 2352-538X
UR  - https://doi.org/10.2991/amitp-16.2016.67
DO  - 10.2991/amitp-16.2016.67
ID  - Li2016/09
ER  -