抄録J-014
Hand shape recognition using position invariant feature from the sequence of depth images
Hand shape recognition using position invariant feature from the sequence of depth images
◎張 燕・Chendra Hadi SURYANTO・福井和広（筑波大）
This paper proposes a system to operate a robot in real time using both-hands shape tracking from a depth camera like SoftKinectic DepthSense 325. In order to achieve this proposal our algorithm makes use of the Kernel Orthogonal Mutual Subspace Method (KOMSM), a kernel-based method for classifying sets of patterns. The system detects the hands shapes and compares the similarity of a sequence of depth images of them with the gestures accepted to control the robot. If the sequence is enough similar with an existing shape, the corresponding command is transmitted to the robot. The main goal of the proposed system is to create a natural and reliable way to control and interact with robots.
