Pedestrian detection with D-CNN

  • Zhonghua Gao
  • , Weiting Chen*
  • , Guitao Cao
  • , Peng Chen
  • *Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

Pedestrian detection plays an important role in intelligent analysis of images and videos. In this paper, we propose a deformation model based convolutional neural network(D-CNN) for pedestrian detection. Enlightened by YOLO model, D-CNN network integrates deformation and occlusion handling into the network to improve the accuracy of occluded pedestrian detection. The performance of D-CNN is evaluated on two popular datasets as well as pictures got in daily life. Among the state-of-the-art methods compared in this paper, the comprehensive performance of D-CNN is the best, whose mAP is only 0.4 points lower than the highest one but the detection speed doubles. So our proposed network can get real-time speed while maintaining rather satisfying precision of pedestrian detection.

Original languageEnglish
Title of host publicationEmbedded Systems Technology - 15th National Conference, ESTC 2017, Revised Selected Papers
EditorsYuanguo Bi, Gang Chen, Qingxu Deng, Yi Wang
PublisherSpringer Verlag
Pages171-180
Number of pages10
ISBN (Print)9789811310256
DOIs
StatePublished - 2018
Event15th National Conference on Embedded Systems Technology, ESTC 2017 - Shenyang, China
Duration: 17 Nov 201719 Nov 2017

Publication series

NameCommunications in Computer and Information Science
Volume857
ISSN (Print)1865-0929

Conference

Conference15th National Conference on Embedded Systems Technology, ESTC 2017
Country/TerritoryChina
CityShenyang
Period17/11/1719/11/17

Keywords

  • D-CNN
  • Deformation handling
  • Occlusion handling
  • Pedestrian detection

Cite this