A Novel Performance Evaluation Methodology for Single-Target Trackers

Researchers from the Faculty of Computer and Information Science of the University of Ljubljana created a rigorous methodology for the analysis of trackers and a methodology for systematic determination of the test database with a large descriptive power, and with the intention to expand the impact of proposed methodologies they established the initiative VOT Challenge, which experienced outstanding support in the community.


Authors:
Matej Kristan, Aleš Leonardis, Luka Čehovin Zajc


Visual tracking of objects is a rapidly evolving field with a wide range of applications that extend from video surveillance systems to autonomous robots. Tens of new approaches are annually published only on high-profile Conferences on Computer Vision, while the lack of a standardized tracker analysis methodology hinders a clear overview of the development of the respective field.

As a due response, the researchers from the Faculty of Computer and Information Science of the University of Ljubljana (Matej Kristan, Luka Čehovin Zajc, Aleš Leonardis), in collaboration with international research groups, carried out an extensive theoretical and experimental study and created a rigorous methodology for the analysis of trackers. The results were published in two of the most prestigious computer and information science journals. To extend the impact of proposed methodologies they established the initiative VOT Challenge (http://www.votchallenge.net), through which they have been organizing challenges and workshops on major Conferences on Computer Vision for the last five years.

The initiative has received tremendous support in the community, while the methodology is becoming the standard for the analysis of trackers. The VOT website registers over 4,000 monthly visits, and the article with the results of the final challenge comprises over one hundred authors who took part in the analysis. The article with the results of the last year’s challenge alone registered over 820 views in one year on the ResearchGate portal.
 

Sources: Kristan, M., Matas, J., Leonardis, A., Vojir, T., Pflugfelder, R., Fernandez, G., Nebehay, G., Porikli, F., Čehovin Zajc, L. (2016) A Novel Performance Evaluation Methodology for Single-Target Trackers, IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI), 38 (11), 2137–2155, (IF = 8.329) (2/133), pure quotes: 14, quotes on Google scholar: 68.
Čehovin Zajc, L., Leonardis, A., Kristan, M. (2016), Visual object tracking performance measures revisited, IEEE Transactions on Image Processing (TIP), 25 (3), 1261–1274, (IF = 4.828) (13/133), pure quotes: 10, quotes on Google scholar: 40.
Kristan, M., Leonardis, A., Matas, J., Felsberg, M., Pflugfelder, R., Čehovin Zajc, L., Vojir, T., Hager, G., Lukežič, A., Fernandez, G., et al. (2016) The Visual Object Tracking VOT2016 challenge results, VOT2016 workshop, Proceedings of ECCV2016, Springer, (IF = 0.552) (151/302), pure quotes: 5, quotes on Google scholar: 384.
Čehovin Zajc, L.. (2017) TraX : the visual Tracking eXchange protocol and library, Neurocomputing, 2017, (260), 5-8.
Čehovin Zajc, L., Lukežič, A., Leonardis, A., Kristan, M., Beyond standard benchmarks: Parameterizing performance evaluation in visual object tracking, International conference on computer vision 2017, ICCV2017, Google scholar CVPR journals and conferences h-5 index: 89, Rate of accepted articles: 28%