Open AccessCross- Scenario Person Re-identification
Author(s) -
Ruoran Jia,
Shuguang Liu
Publication year - 2020
Publication title -
journal of physics. conference series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.21
H-Index - 85
eISSN - 1742-6596
pISSN - 1742-6588
DOI - 10.1088/1742-6596/1684/1/012071
Subject(s) - clutter , computer science , representation (politics) , feature (linguistics) , minimum bounding box , artificial intelligence , bounding overwatch , identification (biology) , task (project management) , domain (mathematical analysis) , modal , object (grammar) , position (finance) , pattern recognition (psychology) , computer vision , image (mathematics) , machine learning , mathematics , engineering , philosophy , systems engineering , law , mathematical analysis , linguistics , chemistry , biology , telecommunications , political science , radar , botany , finance , politics , polymer chemistry , economics
Person Reid is a challenging task for two factors, first one is background interference, such as changes in light, weather, posture, and camera position. Second is domain adaptive capacity, such as model train by market1501 achieve the same performance on Duke dataset. To solve the above problems, we come up with adopt human semantic to remove clutter from unwanted background information, is naturally a better alternative compare with bounding box, we adopt Local Regions Representation to extra the image features, which can preeminently improve the representation of local feature and global feature. Our proposed CSReID integrates human semantic and Local Regions Representation in person re-identification and not need to train on the evaluation dataset can achieve state of the art cross-modal performance.