A Study of Gesture Trajectory Recognition Based on Multi-Time Decision Tree

• Main Authors: CHIH, CHIA-CHIA, 池佳家
• Other Authors: HUANG, SHYH-FANG
• Format: Others
• Language: zh-TW
• Published: 2019
• Online Access: http://ndltd.ncl.edu.tw/handle/m5urtr

碩士 === 明道大學 === 資訊傳播學系碩士班 === 107 === Since the 1995s, researchers have started to get involved in gesture recognition researches. However, their research methods on gesture recognition mostly focused on the differences between the shapes of the palm, or the interpretation of the postures of human bodies. There is no relevant research on the interpretation of the dynamic process of the gesture. Therefore, in this study, we use the relative orientation of the gesture trajectory to encode the dynamic process of the gesture and match the establishment of the multi-time decision tree to achieve instant gesture recognition. The proposed method uses the Kinect hardware to capture the skeleton information and uses the right-hand palm as the input interface in the experiment. The experimental data shows that the proposed method of trajectory encoding, and multi-time decision tree could be utilized to achieve instant recognition of gesture trajectories. The proposed method is an instant and continuous recognition method, that is, the user can input his or her gesture at any time without a specific starting gesture. The results show that the average detection rate is 79.58% on the single-symbol writing recognition project. On the continuous symbol writing recognition project, the average detection rate is 73.14 %, which confirms the feasibility and practicality of this method.