Abstract—Multi-person tracking in videos is a promising but challenging visual task. Recent progress in this field has
introduced deep convolutional features as appearance models, which achieve robust tracking results when coupled with proper motion models. However, model failures that often cause severe tracking problems have not been well discussed and addressed in previous work. In this paper, we propose a solution using online detection of such failures and accordingly adjusting the coupling between appearance and motion models.
The strategy is to let the functional models take over when certain models face data association ambiguity and simultaneously suppress the influence of inappropriate observations during the model update. Experimental results
have proven the benefit of our proposed improvement.
Index Terms—Multiple object tracking, deep neural network,
online learning, tracking-by-detection, multiple hypothesis tracking.
The authors
are with the Université de Lyon, CNRS, INSA-Lyon, LIRIS, UMR5205, F-69621 France
(e-mails: bonan.cuan@insa-lyon.fr, khalid.idrissi@insa-lyon.fr, christophe.garcia@insa-lyon.fr).
[PDF]
Cite: Bonan Cuan, Khalid Idrissi, and Christophe Garcia, "Online Appearance-Motion Coupling for Multi-Person
Tracking in Videos," International Journal of Modeling and Optimization vol. 9, no. 2, pp. 72-76, 2019.
