Boosted Multiple Kernel Learning for First-Person Activity Recognition

摘要

Activity recognition from first-person (ego-centric) videos has recentlygained attention due to the increasing ubiquity of the wearable cameras. Therehas been a surge of efforts adapting existing feature descriptors and designingnew descriptors for the first-person videos. An effective activity recognitionsystem requires selection and use of complementary features and appropriatekernels for each feature. In this study, we propose a data-driven framework forfirst-person activity recognition which effectively selects and combinesfeatures and their respective kernels during the training. Our experimentalresults show that use of Multiple Kernel Learning (MKL) and Boosted MKL infirst-person activity recognition problem exhibits improved results incomparison to the state-of-the-art. In addition, these techniques enable theexpansion of the framework with new features in an efficient and convenientway.