Real-time gesture recognition from dynamic images for construction of an interactive system: Application to a virtual conductor system

✍ Scribed by Takahiro Watanabe; Chil-Woo Lee; Masahiko Yachida

• Publisher: John Wiley and Sons
• Year: 1998
• Tongue: English
• Weight: 207 KB
• Volume: 29
• Category: Article
• ISSN: 0882-1666
• DOI: 10.1002/(sici)1520-684x(199805)29:5<62::aid-scj7>3.0.co;2-l

✦ Synopsis

This paper discusses a method for real-time gesture recognition from a dynamic image and its application to an interactive system. The proposed technique enables a large volume of dynamic image information to be processed efficiently, by using a low-dimensional gesture space to represent a partial image of a specific member performing a gesture. First, a specific member performing a gesture is robustly extracted from the image in real time using a maskable template model (MTM). MTM originates from the conventional template matching technique that was improved by the authors so that it enables soft matching with various deformable objects. Next, the extracted sequence of member images is projected into gesture space and represented as an input gesture curve. The gesture is recognized by comparing this curve with a model gesture curve. The gesture space is constructed in advance by KL expansion of the real-time image series of the model gesture performing member. The authors have used this technique to construct a real-time interactive systemvirtual conduc-tor system. This system recognizes conducting gestures performed by a user for real-time control of music played by a computer. The system demonstrated the efficiency of the disclosed technique.