Moving Pedestrian Detection Using Normed Proposals and Key Points Matching

DOI

10.2991/icmemtc-16.2016.124How to use a DOI?

Keywords

pedestrian detection; region of interest; key point; proposal.

Abstract

Occlusion detection and automatic adaption of a generic pedestrian detector to a specific scene are difficult problems in intelligent monitoring. When a detector trained in a specific scene is applied on a new scene, its accuracy will decrease greatly. To solve this problem, we propose a new detection algorithm in which motion regions of interest based on motion information are obtained quickly by a flash-bit computing method. Also we focus on the case in which a single target converts to be a difficult one due to multiple overlapping between pedestrians. Key points with BRISK feature which computed and saved before are used to match difficult targets in occlusions. Normed proposals which proved to have higher confidence are used to correct the location and shape of detection windows, results in a five percent increasing of detection accuracy. Results of comparative experiments of five different detectors on three motion pedestrian datasets show that proposed algorithm achieves not only a real time speed, but also the best accuracy that more than half of difficult targets are detected successfully.

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  - Wei Chen
AU  - Taihong Wang
AU  - Yong Cai
PY  - 2016/04
DA  - 2016/04
TI  - Moving Pedestrian Detection Using Normed Proposals and Key Points Matching
BT  - Proceedings of the 2016 3rd International Conference on Materials Engineering, Manufacturing Technology and Control
PB  - Atlantis Press
SP  - 627
EP  - 635
SN  - 2352-5401
UR  - https://doi.org/10.2991/icmemtc-16.2016.124
DO  - 10.2991/icmemtc-16.2016.124
ID  - Chen2016/04
ER  -