BAG OF WORDS APPROACH AND DOCUMENT-TOPIC MODELING FOR HUMAN ACTIVITY RECOGNITION FROM VIDEOS

Janson Hendryli

Abstract


Human activity recognition from videos have many useful real world applications, ranging from multimedia, entertainment, and security. In this paper, an approach inspired by a popular text document, namely the bag of words and document topic modeling, is explored. The latent Dirichlet allocation (LDA) and non-negative matrix factorization (NMF) are used to model the latent topic distribution in videos. Finally, the discovered distribution can be used to transformed the bag of words representation in order to categorize the video into ten daily human activities. The classification is done by feeding the transformed term-frequency of the visual words to the logistic regression and SVM model. The NMF achieved higher F1-score than the LDA when both SVM and logistic regression is used as the classifier.

Keywords: human activity recognition, bag of words, document topic modeling

Full Text:

PDF

Refbacks

  • There are currently no refbacks.