%0 Conference Proceedings
%T Deep Visible and Thermal Image Fusion with Cross-Modality Feature Selection for Pedestrian Detection
%+ Collège of information Engineering (CIE)
%+ Beijing Key Laboratory of Light Industrial Robot and Safety Verification
%+ Advanced Innovation Center for Imaging Technology
%A Li, Mingyue
%A Shao, Zhenzhou
%A Shi, Zhiping
%A Guan, Yong
%Z Part 2: AI
%< avec comité de lecture
%( Lecture Notes in Computer Science
%B 17th IFIP International Conference on Network and Parallel Computing (NPC)
%C Zhengzhou, China
%Y Xin He
%Y En Shao
%Y Guangming Tan
%I Springer International Publishing
%3 Network and Parallel Computing
%V LNCS-12639
%P 117-127
%8 2020-09-28
%D 2020
%R 10.1007/978-3-030-79478-1_10
%K Pedestrian detection
%K Cross-modality features
%K Feature fusion
%Z Computer Science [cs]Conference papers
%X This paper proposes a deep RGB and thermal image fusion method for pedestrian detection. A two-branch structure is designed to learn the features of RGB and thermal images respectively, and these features are fused with a cross-modality feature selection module for detection. It includes the following stages. First, we learn features from paired RGB and thermal images through a backbone network with a residual structure, and add a feature squeeze-excitation module to the residual structure; Then we fuse the learned features from two branches, and a cross-modality feature selection module is designed to strengthen the effective information and compress the useless information during the fusion process; Finally, multi-scale features are fused for pedestrian detection. Two sets of experiments on the public KAIST pedestrian dataset are conducted, and experimental results show that our method is better than the state-of-the-art methods. The robustness of fused features is improved, and the miss rate is reduced obviously.
%G English
%Z TC 10
%Z WG 10.3
%2 https://inria.hal.science/hal-03768739/document
%2 https://inria.hal.science/hal-03768739/file/511910_1_En_10_Chapter.pdf
%L hal-03768739
%U https://inria.hal.science/hal-03768739
%~ IFIP-LNCS
%~ IFIP
%~ IFIP-TC
%~ IFIP-TC10
%~ IFIP-NPC
%~ IFIP-WG10-3
%~ IFIP-LNCS-12639