Loading...

4D Human Action Recognition Using A Fixed RGB-D Camera

Khatami Nejad Tehrani, Ahmad | 2016

670 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 48838 (19)
  4. University: Sharif University of Technology
  5. Department: Computer Engineering
  6. Advisor(s): Kasaei, Shohreh
  7. Abstract:
  8. Human action recognition is one of the computer vision branches. Video surveillance and human-computer interaction is among its modern applications. The main goal of this subject is to label RGB-D videos which are captured from acting human. Therefore, labeling the input videos among pre-learned action is called as action recognition.The action recognition problem consists of two primary parts. The first part is selecting a suitable descriptor to describe input videos, and the other part is the way that the system has been learned based on the learning action data. The main goal of this research title is to propose a method for recognizing each action video (that is acted by human and captured by a camera which has ability to record RGB-D images).Due to the way that human recognizes each action by focusing on the joints which has noticeable movements, proposed method is designed to work based on this learning idea. Proposed method can discriminate between each joint based on its activity. Applying the idea of active joints on action recognition, learning section can cause some unpredictable problems that related to various length of feature vector of each video. Therefore, as part of proposed method, two structure is defined, one for learning and the other for testing.Using active joints can reduce complexity of time and computation. Furthermore, the method that is based on active joints increases the speed and performance of action recognition. This algorithm can also handle the occlusion of body joints in “motion capture” problem & ths loss of inactive joints in “action recognition” problem
  9. Keywords:
  10. Human Action Recognition ; Human Activity Recognition ; RGB-D Camera ; Hidden Markov Model ; Human Body Skeleton Model ; Four Dimensional Recognition

 Digital Object List

 Bookmark

No TOC