Kinect depth video compression for action recognition

Since the advent of the Kinect camera, depth videos have become easily accessible to consumers and researchers, allowing a variety of complex classification tasks to be done more accurately and easily than with RGB videos. The wide use of Kinect has created a need for effective compression algorithms. We present three compression schemes, all evaluated using a classification metric for human activity recognition. The first scheme uses the idea of companding to pre-process the data prior to compressing it with a standard H.264 coder. The second scheme uses a standard H.264 coder and appends additional feature bits to the compressed signal to aid in classification. The third compression scheme also uses a standard H.264 coder and attempts to improve classification performance by learning a mapping between features extracted from compressed videos and features extracted from uncompressed videos.

Name thesis:degree_name

M.S.

Level thesis:degree_level

Thesis

Discipline thesis:degree_discipline

Electrical & Computer Engr

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2014

Author dc:creator

• Fedorov, Igor

Contributors dc:contributor

• Moulin, Pierre

• Kinect
• Depth
• Video
• Compression
• Action
• Recognition

Statement dc:rights

• Copyright 2014 Igor Fedorov

Language dc:language

en

Handle dc:identifier

http://hdl.handle.net/2142/49462

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/49462