Short and long-term tracking
We consider online tracking, defined as following the location of one target in video sequences, starting from no more object-specific knowledge than the bounding box in the starting frame. Tracking is a very hard problem as the target appearance and model is unknown, the illumination, reflection and motion are irregular, the pose, scale and view frequently change, and the background or foreground may interfere. All these aspects have to be learned from the first frame or when they happen. One immediate consequence is that large datasets with the emphasis on variety of the data are needed to assess the quality of tracking solutions. A new dataset, ALOV300++ was generated for that purpose. The emphasis there is on short sequences as coverage of variety where errors appear to occur usually within a ten-frame interval appeared to be most useful. In a collaborative study [Smeulders PAMI 2014] with Orlando and Modena, 19 current trackers have been evaluated in depth. Their results as well as the driving mechanisms will be discussed. In the last part of the talk we reflect on the consequence for long-term tracking. In what way does it differ from short tracking? Does long-term require a different type of testing and different evaluation measures? Will the era of large-scale concept recognition redefine tracking? One could pose online tracking as a generic image search problem, e.g. [Ran Tao CVPR 2014], looking for an instance?
1. Arnold W. M. Smeulders, Dung M. Chu, Rita Cucchiara, Simone Calderara, Afshin Dehghan, Mubarak Shah: Visual Tracking, an Experimental Survey, IEEE Transactions on Pattern Analysis and Machine Intelligence, 2014.
2. Ran Tao, Efstratios Gavves, Cees Snoek, Arnold Smeulders: Locality in Generic Instance Search from One Example, CVPR 2014.
Arnold W.M. Smeulders graduated from Delft Technical University in Physics and Leyden Medical Faculty (PhD) both on a computer vision topic of the time. He is professor at the University of Amsterdam, leading the ISIS group on visual search engines. The search engine has been a top three performer in the TRECvid competition for the last 10 years. In 2010 he co-founded Euvision a university spin-off. He is scientific director of COMMIT/, the large public private-ICT-research program of the Netherlands. He is past associated editor of the IEEE trans PAMI and current of IJCV. He is IAPR fellow, Member of the Academia Europeana. He was visiting professor in Hong Kong, Tsukuba, Modena, Cagliari, and Orlando.