[Next][Up][Previous]
Next:ReferencesUp:Virtual Kathakali: Gesture-Driven Previous:Reproduction of User's

Conclusions

In this paper, we have presented a communication environment in which a person could transform himself into any character he desires in a virtual environment. In our case, the character was that of a Kathakali Dancer. Computer Vision techniques are used for passive detection of user's arm pose in real time. Some simple constraints on the user (that he stand in front of a dark background, and that his skin tone differs from his clothing) result in being able to use some extremely simple and fast detection methods based on dynamic threshold binarization. A single camera is used and the user does not need to wear any special devices. The 3-D model is instantiated during the calibration phase, and the user's motions are reproduced using various transformations on the dancer model. Our implementation showed good performance with a processing speed of 5 frames per seconds on a very old and simple image processing system.

The primary application for such a system would be in the coaching of motor skills with distant trainers and coaches. In the arts, such a system would permit quick informal recording of creative insights, which would otherwise require elaborate stage and makeup arrangements before being presented in the final form. In general, the expansion of Internet, even with low-bandwidth connectivity, will permit gestural interaction of this type as long as major computational tasks such as identification of limb postures and recreation/graphics output are performed on the local machines. In the coming decades, systems of this type are likely to have a profound effect on the declining trend in artistic traditions worldwide. Also, in sports as well as critical operations such as surgery, this could enable expert trainers to provide guidance to a far larger set of students than would be possible in a face-to-face mode.

Since the image processing is carried on an image in 2D, the system is not able to resolve between two arm poses with the same projection, as when the entire arm is on a horizontal plane. However, due to the very nature of this deficiency, it will not matter to the viewer so long as it is viewed from the same angle.

In this work, we have only used the arm pose of the user to control the virtual model. In traditional dance form, facial expressions and finger movements constitute an important component of the dancer's emotional expression ( abhinaya). In the current phase, with a single camera of fixed resolution, this is not possible; in fact, no vision system today can model both the finger and the gross body motions. However, some beginnings have been made towards integrating face recognition by having a camera look down on the user's face from a fixture mounted on the head itself. With multiple cameras, one camera could be used for the full-body field of view, whereas one more other cameras could track the important aspects of the dance - particularly the two hands and also the face. Such a multi-scale imaging and tracking system would enable detailed reconstructions of the important aspects of the scene.

Moreover the system, which supports only a single user at present, can easily be extended to support multiple users all sharing the same virtual environment. Hence creation of Virtual Theaters become a distinct possibility where different actors situated far away from each other can be actors in the same virtual theater. Since the bandwidth required by the system is very small, transmitting the data over the Internet is a feasible option. Another challenging problem is to have real and virtual actors share the same space in a seamless manner from the audience's viewpoint.
 


[Next][Up][Previous]
Next:ReferencesUp:Virtual Kathakali: Gesture-Driven Previous:Reproduction of User's