Visual object tracking is an important problem of Computer Vision that aims to estimate the object position given an initial bounding box of the object of interest in the first frame of a video sequence. Recent methods focus on the design of deep network architectures to increase the tracking accuracy and/or computational efficiency. Correlation filter based methods [1,2] along with Siamese deep network models  are now popular due to the fact that they can densely sample negative and positive examples in the online/offline phase of the tracking. In our recent work, we have shown that training a Siamese network to decrease the correlation filter based tracking loss significantly increases the tracking performance and won the winner position in Visual Object Tracking Challenge 2017 .
Even though some efforts have been put on bounding box regression in the context of Siamese tracking framework , the key challenges of object tracking, such as severe rotation, small objects and very high or low aspect ratio in the bounding box of the objects, have not been directly addressed in a unified learning framework since the efficient approaches are not suitable to precisely consider these challenges by design. Moreover, short-term tracking methods of the literature do not consider the cases when the object is not visible for a while.
In this project, the candidate will develop methods that take into account those challenges and design an end-to-end learning framework which is capable of operating in real-time. We would like to exploit the recent advancements in the object detection literature to increase the tracking accuracy. Lastly, we would like to concentrate on long-term object tracking methodologies to re-detect and accurately track the target object in the case of full occlusions or temporary invisibility of the objects.
 E. Gundogdu and A. A. Alatan, "Good Features to Correlate for Visual Tracking," in IEEE Transactions on Image Processing, 2018
 L. Bertinetto, J. Valmadre, J. F. Henriques, A. Vedaldi, and P. H. S. Torr. "Fully-convolutional siamese networks for object tracking." ECCV, 2016.
 J. Valmadre, L. Bertinetto, J. F. Henriques, A. Vedaldi, and P. H. Torr, "End-to-end representation learning for correlation filter based tracking," arXiv preprint, 2017.
 M. K. et.al., “The visual object tracking vot2015 challenge results,” in ICCV Workshops, 2017.
 Li, Bo and Yan, Junjie and Wu, Wei and Zhu, Zheng and Hu, Xiaolin, "High Performance Visual Tracking With Siamese Region Proposal Network, CVPR, 2018
 M. Danelljan, A. Robinson, F. Shahbaz Khan, and M. Fels- berg. Beyond correlation filters: Learning continuous convolution operators for visual tracking. In ECCV, 2016.
Back to the project list.
The candidate should have programming experience, ideally in Python. Previous experience with machine learning and computer vision is a plus.
30% Implementation, 30% Theory, 40% Research and Experiments