Abstract | ||
---|---|---|
Learning to recognize pedestrian attributes (such as gender, hair style, take hat or not) in video surveillance scenarios is critical to a variety of tasks, such as crime prevention and border control. However, it is still challenging due to low resolution and highlight influence in the actual surveillance scenarios, in which traditional methods work not well. This paper aims at proposing a robust pedestrian attribute recognition framework which can be adaptive to the actual surveillance scenarios. Specifically, we first propose a hierarchical recognition strategy by heuristically classifying the pedestrian attributes as global ones (such as gender and age) and local ones (such as hair style and has glass). Then the whole region is used for the global attribute recognition, and the relevant regions are used for the local attribute recognition. To estimate the relevant regions above, we further propose an adaptive region localization scheme, including position estimation based on geometric human body and relevant region localization based on random expansion. Finally, experimental results on representative datasets and our actual surveillance scenarios both demonstrate the effectiveness of the proposed method. |
Year | DOI | Venue |
---|---|---|
2017 | 10.1109/ICMEW.2017.8026292 | 2017 IEEE International Conference on Multimedia & Expo Workshops (ICMEW) |
Keywords | Field | DocType |
CNN,Adaptive region localization,Pedestrian attribute,Hierarchical | Computer vision,Data mining,Pedestrian,Heuristic,Computer science,Artificial intelligence,Machine learning,Crime prevention | Conference |
ISSN | ISBN | Citations |
2330-7927 | 978-1-5386-0561-5 | 0 |
PageRank | References | Authors |
0.34 | 16 | 4 |
Name | Order | Citations | PageRank |
---|---|---|---|
Chunfeng Yao | 1 | 0 | 0.68 |
Bailan Feng | 2 | 9 | 3.14 |
Defeng Li | 3 | 0 | 0.34 |
Jian Li | 4 | 162 | 44.60 |