Human Activity Recognition with Spatio Temporal Features in RGB-D Videos

Please enable javascript in your browser.

Ebtehaj, Ali | 2015

545 Viewed

Type of Document: M.Sc. Thesis
Language: Farsi
Document No: 47579 (19)
University: Sharif University of Technology
Department: Computer Engineering
Advisor(s): Jamzad, Mansour
Abstract:
Human activity recognition is an important and useful area in computer vision that application include surveillance systems, patient monitoring systems, human-computer interaction and analyse video data from big websites.Traditional Human action recognition use the RGB videos as default input that unable describe motion and action as full. On the other hand Kinect camera sendsthe RGB data to output in addition to the Depth Data that allows us to extract skeleton of human easily. Recently Space-time features have been particulary popular in RGB Videos because of their structure. These features are describedby their descriptor and send the good and important information to output.Finally we concluded that we can improve the performanc if we use the space time and skeleton features extracted from RGB and Depth data. Our method use the Bag Of Words model and fisher vector encoding on the mentioned features. Itimproved both the accuracy and performance compared with the most recent methods
Keywords:
Locational-Temporal Feature ; Words Bag Model ; Kinect Sensor ; RGB-D Camera ; Human Activity Recognition ; Fisher Vector

No TOC