Loading...

Dynamic Hand Gesture Recognition Using Bag of Words

Amirkhani Najafabadi, Mohammad | 2014

843 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 46261 (19)
  4. University: Sharif University of Technology
  5. Department: Computer Engineering
  6. Advisor(s): Rabiee, Hamid Reza
  7. Abstract:
  8. New applications of gesture recognition problem have emerged that are constrained by lack of sufficient training data. Bag-of-words methods are attractive in this type of scenario and have come to attention again. A typical bagof-words system has four stages: feature extraction, dictionary learning and feature coding, pooling, and classification. The dictionary learning and feature coding stage is important due to its effect on classifier complexity and the discriminative capability of the features that affect the final result. Existing methods in gesture recognition have neglected the relation between different gestures when learning a dictionary. In this thesis, we focus on the relations between feature vectors of different classes and propose a method that constructs
    a discriminative dictionary. Since training data is scarce, atoms are selected from the set of available features based on their discriminative power and their ability to reconstruct features from the same gesture. Hence a major benefit of our method is its ability to handle lack of training data and yet construct a dictionary with both reconstructive and discriminative capabilities. The proposed method has been Experimentally evaluated and compared against other gesture recognition methods on one-shot data sets. Experimental results show
    numerically and visually that our proposed method achieves substantial gains in comparison to the state-of-the-art, namely 6 % reduction in the Levenshtein distance with negligble computational overhead
  9. Keywords:
  10. Movement Recognition ; Sparse Representation ; Locational-Temporal Feature ; Dictionary Learning ; Hand Gestures ; Discriminative Dictionary

 Digital Object List

 Bookmark

...see more