Deep Occlusion Reasoning for Multi-Camera Multi-Target Detection

被引:53
作者
Baque, Pierre [1 ]
Fleuret, Francois [1 ,2 ]
Fua, Pascal [1 ]
机构
[1] Ecole Polytech Fed Lausanne, CVLab, Lausanne, Switzerland
[2] IDIAP, Martigny, Switzerland
来源
2017 IEEE INTERNATIONAL CONFERENCE ON COMPUTER VISION (ICCV) | 2017年
基金
瑞士国家科学基金会;
关键词
TRACKING;
D O I
10.1109/ICCV.2017.38
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
People detection in single 2D images has improved greatly in recent years. However, comparatively little of this progress has percolated into multi-camera multi-people tracking algorithms, whose performance still degrades severely when scenes become very crowded. In this work, we introduce a new architecture that combines Convolutional Neural Nets and Conditional Random Fields to explicitly model those ambiguities. One of its key ingredients are high-order CRF terms that model potential occlusions and give our approach its robustness even when many people are present. Our model is trained end-to-end and we show that it outperforms several state-of-the-art algorithms on challenging scenes.
引用
收藏
页码:271 / 279
页数:9
相关论文
共 29 条
[1]  
Alahi A., 2011, J MATH IMAGING VISIO
[2]  
[Anonymous], 2016, Variational inference: A review for Statisticians
[3]  
[Anonymous], P 13 INT C ART INT S
[4]  
Arnab A., 2015, CORR
[5]   Social Scene Understanding: End-to-End Multi-Person Action Localization and Collective Activity Recognition [J].
Bagautdinov, Timur ;
Alahi, Alexandre ;
Fleuret, Francois ;
Fua, Pascal ;
Savarese, Silvio .
30TH IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR 2017), 2017, :3425-3434
[6]  
Baque P., 2016, C COMP VIS PATT REC
[7]  
Baque P., 2017, C COMP VIS PATT REC
[8]   Multiple Object Tracking Using K-Shortest Paths Optimization [J].
Berclaz, Jerome ;
Fleuret, Francois ;
Tueretken, Engin ;
Fua, Pascal .
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, 2011, 33 (09) :1806-1819
[9]  
Chavdarova T., 2017, ABS170204593 ARXIV
[10]  
Domke J., 2013, CORR