2010
DOI: 10.1016/j.patcog.2009.05.015
|View full text |Cite
|
Sign up to set email alerts
|

Action categorization with modified hidden conditional random field

Abstract: In this paper, we present a method for action categorization with a modified hidden conditional random field (HCRF). Specifically, effective silhouette-based action features are extracted using motion moments and spectrum of chain code. We formulate a modified HCRF (mHCRF) to have a guaranteed global optimum in the modelling of the temporal action dependencies after the HMM pathing stage. Experimental results on action categorization using this model are compared favorably against several existing modelbased m… Show more

Help me understand this report

Search citation statements

Order By: Relevance

Paper Sections

Select...
3
1
1

Citation Types

0
25
0

Year Published

2010
2010
2018
2018

Publication Types

Select...
7
2

Relationship

0
9

Authors

Journals

citations
Cited by 68 publications
(25 citation statements)
references
References 24 publications
0
25
0
Order By: Relevance
“…This allows for the modeling of complex interactions between labels and long-range dependencies, while inference is approximate instead of exact as in CRFs. Zhang and Gong [178] use a hidden CRF (HCRF, [111]) to label sequences as a whole. They introduce a HMM pathing stage, which ensures that learning the HCRF parameters is globally optimal.…”
Section: Discriminative Modelsmentioning
confidence: 99%
“…This allows for the modeling of complex interactions between labels and long-range dependencies, while inference is approximate instead of exact as in CRFs. Zhang and Gong [178] use a hidden CRF (HCRF, [111]) to label sequences as a whole. They introduce a HMM pathing stage, which ensures that learning the HCRF parameters is globally optimal.…”
Section: Discriminative Modelsmentioning
confidence: 99%
“…(51) with the same initialization (45), and backwards recursion (43). This construction gives, in turn, rise to another issue: what is the appropriate selection of the values {ŷ τ } t−2 τ =1 ?…”
Section: Sequence Decodingmentioning
confidence: 99%
“…For example, in natural language tasks, useful features include neighboring words and word bigrams, prefixes and suffixes, capitalization, membership in domainspecific lexicons, and semantic information from sources such as WordNet [36]. During the last years, we have witnessed an explosion of interest in CRFs, as it has managed to achieve superb prediction performance in a variety of scenarios, thus being one of the most successful approaches to the structured output prediction problem, with successful applications including text processing, bioinformatics, natural language processing, and computer vision [21], [13], [20], [27], [35], [45], [26], [29].…”
Section: Introductionmentioning
confidence: 99%
“…Researchers who used silhouettes have followed one of two major categories to classify actions. One is to extract patterns from silhouette sequence [4,5,6,7,14,15] and the other is to create a model from each silhouette [6,8,9,10,11,16,17]. The major challenges for human action recognition using 3D skeletal joint locations are: selection of significant frames, overlapping actions, diversity in actions, representation of temporal sequence, arbitrary viewing angel, discriminative features, noise, subjective interpretation of actions, etc.…”
Section: Introductionmentioning
confidence: 99%