In this paper, we propose a gesture modeling system based on computer vision in order to recognize a gesture naturally without any trouble between a system and a user using real-time 3D modeling information on multiple objects. It recognizes a gesture after 3D modeling and analyzing the information pertaining to the user's body shape in stereo views for human movement. In the 3D-modeling step, 2D information is extracted from each view by using an adaptive color difference detector. Potential objects such as faces, hands, and feet are labeled by using the information from 2D detection. We identify reliable objects by comparing the similarities of the potential objects that are obtained from both the views. We acquire information on 2D tracking from the selected objects by using the Kalman filter and reconstruct it as a 3D gesture. A joint of each part of a body is generated in the combined objects. We experimented on ambiguities using occlusion, clutter, and irregular 3D gestures to analyze the efficiency of the proposed system. In this experiment, the proposed gesture modeling system showed a good detection and a processing time of 30 frames per second, which can be used in a real-time. © Springer-Verlag Berlin Heidelberg 2007.
CITATION STYLE
Kang, B. D., Eom, J. S., Kim, J. H., Kim, C. S., Ahn, S. H., Shin, B. J., & Kim, S. K. (2007). Human motion modeling using multivision. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 4552 LNCS, pp. 659–668). Springer Verlag. https://doi.org/10.1007/978-3-540-73110-8_72
Mendeley helps you to discover research relevant for your work.