Fast and reliable hand action recognition
Metadata[+] Show full item record
[ACCESS RESTRICTED TO THE UNIVERSITY OF MISSOURI AT AUTHOR'S REQUEST.] In this work, we develop a hand action recognition method using a SVM (Support Vector Machine) classifier with HOG (Histogram of Oriented Gradients) features and Motion Vectors. Hand gesture recognition system analyzes the HOG feature using SVM. Hand action recognition system applies motion estimation to the input video, analyze the motion vectors, and then recognize the action using a SVM classifier. Our gesture recognition results show that this method is relatively insensitive to variations in illumination, camera perspective, and background variations. We tested our method on 10000 real life images, which captured on camera under different backgrounds and lighting conditions. We achieved a recognition rate of 94%. In the second part of this thesis, we focus on hand action recognition from videos. Background subtraction is used to obtain the foreground of moving objects. Conceptually, this recognition method is based on motion estimation, searches the block in the current frame, and finds the best match of it in the previous frame. Our hand action recognition results show that 74% of the actions can be successfully recognized.
Access is limited to the University of Missouri--Columbia.