Fingertip Positioning and Tracking Using Double-Layer Particle Filtering with Motion-Driven Mechanism

• Main Authors: Jin-De Chen, 陳晉德
• Other Authors: Shih-Shinh Huang
• Format: Others
• Language: zh-TW
• Published: 2015
• Online Access: http://ndltd.ncl.edu.tw/handle/17925689670281711311

碩士 === 國立高雄第一科技大學 === 電腦與通訊工程研究所 === 103 === The paper proposes an algorithm that uses double-layer particle filtering for positioning and tracking fingertip. The particle with the largest weight among those in the same group is kept in the higher layer. Only the high-layer particles are drifted to local mode with mean-shift algorithm so as to significantly reduce the computation complexity. In addition, the particles organized in such a hierarchical manner will preserve high diversity and well approximate the target probability distribution as compared with traditional particle filtering. Besides, the data-driven mechanism based on motion detection is imposed to draw the samples at the center of change region in the observed image. This makes the system have ability to detect the newly introduced fingertip or resume tracking process in case of losing tracking. In our experiment, the results exhibit that the proposed method significantly improve the efficiency without sacrificing the positioning accuracy.