doi:10.1016/S0167-8655(99)00134-8
Copyright © 2000 Elsevier Science B.V. All rights reserved.
User-independent online gesture recognition by relative motion extraction
a Department of Informatics and Electronics, Tsukuba University, Tsukuba, Japan
Adaptive Vision Lab, Machine Understanding Division, ETL, 1-1-4 Umezono, Tsukuba 305 8568, Japan
1
Received 29 March 1999;
Revised 21 September 1999.
Available online 23 December 1999.
References and further reading may be available for this article. To view references and further reading you must
purchase this article.
Abstract
We propose a new method for user-independent gesture recognition from time-varying images. The method uses relative motion-dependent feature extraction, together with discriminant analysis and dynamically updated buffer structures for providing online learning/recognition abilities. Efficient and robust extraction/representation of information about motion is achieved. Being computationally inexpensive the method allows real-time performance.
Author Keywords: Author Keywords: User-independent online gesture recognition; Human–computer interface (HCI); Relative motion extraction; Discriminant analysis; Machine learning
Fig. 1. General outline of the system.
Fig. 2. Feature extraction mechanism (shown for the gesture “point to the left” for θ=45
, l=20, K=2, X=2; see text for details). All possible binary transition patterns [p1(t), p2(l,nθ,t), p1(t+1), p2(l,nθ,t+1)] for different values of nθ and l are extracted and integrated, as the circular areas are shifted across the sequential images B(t) and B(t+1).
Fig. 3. Gesture trajectories in discriminant feature space for four classes of gestures. Class averages for each class are displayed by solid lines, while trajectories of the corresponding test samples (which have not been used during the training process) are shown in dashed lines.
Fig. 4. Information processing flow performed by the dynamic buffer structures (DBS). Details are given in the text.
Fig. 5. Snapshots of some of the gestures used in Experiment 3.1 (MMDB database). The following gestures are shown: “me” (top), “left circle” (middle) and “expand” (bottom).
Fig. 6. Snapshots of some of the gestures used in Experiment 3.2(b) (different subjects). The following gestures are shown: “clap hands” (top), “cross” (middle) and “banzai” (bottom).
Fig. 7. Snapshots of some of the gestures used in Experiment 3.2(c). The gesture “right hand up” is shown in three different background and illumination conditions.