Local Space-Time Representations for Human Action Recognition: Shugao Ma (PhD Oral Exam)

Starts:
11:00 am on Friday, September 13, 2013
Ends:
12:00 pm on Friday, September 13, 2013
Location:
MCS 148
Abstract: Human action recognition has been an important topic of interest in computer vision, due to its wide ranging application in automatic video analysis, video retrieval and more. Most action recognition systems use machine learning techniques to learn action models, which usually require suitable video representations so that features can be extracted from the representation and fed into the learning system. Holistic representations have been explored, which contain whole video frames or whole human body, such as motion history images. In recent years, many local space-time representations have been proposed, which contain space-time fragments of the video. Using these representations, state-of-art action recognition performance has been achieved on many standard datasets. This talk will introduce the most successful local space-time representations, while making contrast to some classical holistic representations. Committee Members: Stan Sclaroff Margrit Betke George Kollios