Robust incremental hidden conditional random fields for action recognition (Master thesis)
Human action recognition is a challenging topic of computer vision research and continues to receive a keen interest due to the variety of applications that can be used. The creation of a supervised system able to understand and automatically recognize low-level actions and high-level activities is the core problem that these applications attempt to solve. A promising probabilistic graphical model that has been recently proposed for the recognition task is Hidden Conditional Random Fields (HCRF). However, the number of hidden variables that the model incorporates remains a severe limitation of the HCRF due to the fact that the user is asked to make an advance and intuitive assumption for this parameter. In this thesis, we address this limitation by proposing a new model, called Robust Incremental Hidden Conditional Random Fields (RI-HCRF), which estimates the number of hidden states incrementally. Multiple Hidden Markov Models (HMM) are created whose parameters are defined by the potentials of the original HCRF graph. Starting from a small number of hidden states and increasing their number incrementally, the Viterbi path is computed for each HMM. The method seeks for a sequence of hidden states, where each variable participates in a maximum number of optimal paths. Therefore, variables with low participation in optimal paths are rejected. In addition, a robust mixture of Student’s t-distributions is imposed as a regularizer to the parameters of the model. The proposed method is tested in six publicly available datasets using different feature representations. The a priori knowledge of the optimal number of hidden variables and the t-distributed parameters lead to a more robust estimation framework for the classification task. The experiment results show that RI-HCRF estimates successfully the number of hidden states and outperforms all state-of-the-art models that were used as baseline.
|Institution and School/Department of submitter:||Πανεπιστήμιο Ιωαννίνων. Σχολή Θετικών Επιστημών. Τμήμα Μηχανικών Η/Υ & Πληροφορικής|
|Subject classification:||Action recognition|
|Keywords:||Αναγνώριση δραστηριότητας,Κρυφά υπό συνθήκη πεδία,Γραφικό μοντέλο,Action recognition,Hidden conditional random fields,Graphical model|
|Appears in Collections:||Διατριβές Μεταπτυχιακής Έρευνας (Masters)|
Files in This Item:
|Μ.Ε. ΜΑΣΤΟΡΑ ΕΡΜΙΟΝΗ 2017.pdf||5.51 MB||Adobe PDF||View/Open|
Please use this identifier to cite or link to this item:This item is a favorite for 0 people.
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.