Reconocimiento en-línea de acciones humanas basado en patrones de RWE aplicado en ventanas dinámicas de momentos invariantes

2Citations
Citations of this article
18Readers
Mendeley users who have this article in their library.

Abstract

This paper presents a methodology for online human action recognition on video sequences. It addresses an efficient approach to use invariant moments as image descriptors, applied in processing silhouettes obtained from depth maps. A quick comparison between size-4 windows (equivalent to 4 frames) is performed by computing the Mahalanobis distance, on one of the invariant moment sequences identified as less sensitive to noise and more stable during movement absence. This approach is used for rapid detection of the idle/motion state, which allows the capture of dynamic growth intervals (windows) for further processing, rescuing from the signal contained their temporal and frequential properties. By applying the Haar wavelet transform, three decomposition levels are used for calculating RelativeWavelet Energy (RWE - RelativeWavelet Energy) and SSC (Slope Sign Change), obtaining 11-dimensional patterns. In experiments, 97% of 4 movements online-captured were recognized correctly, and 10 movements taken from Muhavi-MAS database were recognized with 94.2% efficiency. © 2014 CEA. Publicado por Elsevier España, S.L. Todos los derechos reservados.

Cite

CITATION STYLE

APA

López, D. R., Neto, A. F., & Bastos, T. F. (2014). Reconocimiento en-línea de acciones humanas basado en patrones de RWE aplicado en ventanas dinámicas de momentos invariantes. RIAI - Revista Iberoamericana de Automatica e Informatica Industrial, 11(2), 202–211. https://doi.org/10.1016/j.riai.2013.09.009

Register to see more suggestions

Mendeley helps you to discover research relevant for your work.

Already have an account?

Save time finding and organizing research with Mendeley

Sign up for free