Téma:Learnable predictors for detection and tracking in images and videos
Vedoucí:Prof. Ing. Tomáš Svoboda Ph.D.
Vypsáno jako:Diplomová práce,Bakalářská práce,Individuální projekt,Dobrovolná odborná práce
Popis:We will follow the idea of learnable predictors/detectors. Simply
speaking the principle is to collect a few training images and learn a
direct mapping between observations and parameters, e.g. motion. The
learning process explicitly optimizes the predictor complexity
w.r.t. predefined accuracy and operating range (basin of attraction)
The approach proved to be extremely efficient and robust for objects
whose appearance were available for off-line learning in advance. We
focus on paradigm person generic detector by appearance encoded
regression. The learning procedure will separate appearance variations
in unsupervised manner. The approach will allow for learning on few
class examples (people) and detection/tracking on the complete class
Literatura:[1] K. Zimmermann, J. Matas, and T. Svoboda. Tracking by an Optimal Sequence of Linear Predictors. IEEE Transactions on Pattern Analysis and Machine Intelligence. 2008
[2] K. Zimmermann, T. Svoboda, J. Matas. Simultaneous learning of motion and appearance, ECCV 2008 Workshop on Machine Learning for Vision-based Motion Analysis
Vypsáno dne:12.12.2009