Open AccessA Study of Community Surveillance System Improvement based on ResNet Person Re-identification
Author(s) -
Xueping Gu,
Chengzhang Qu
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 - 10.1088/1742-6596/1575/1/012231
Subject(s) - computer science , classifier (uml) , identification (biology) , artificial intelligence , rank (graph theory) , data mining , norm (philosophy) , computer vision , machine learning , pattern recognition (psychology) , mathematics , botany , combinatorics , biology , political science , law
Most existing community surveillance systems rely on face recognition to identify pedestrian. The face image captured by community surveillance camera are not always clear enough. Thus, Person re-identification (ReID) imply critical applications in surveillance as it has more image information of pedestrian. In this paper, we refine a ResNet50 based reID model which only adds a Linear layer, a Batch Norm layer and a reLU layer in front of the classifier. The refined model is simple to build on the surveillance system, and we have tested in our demo surveillance system and Market1501 data-set. The experiment result shows it can work well on real-time. On Market1501, our results are competitive that rank-1=0.875000, rank-5=0.944477, and mAP=0.706647 since its low complexity. Our demo community surveillance system shows the refined ReID model is competent and practical for identification tasks.