我們提出一個基於模型的方法,能夠從使用單一攝影機拍攝的影像序列中,估計出影像裡手部的姿勢。這個方法不需要設置標記(marker),也就是不需要在手上設置標記來輔助估測。我們會計算一些特徵,包含目標函數(objective function)會用到的剪影(silhouette)以及邊緣(edge)等。這些特徵是分別從輸入影像,以及利用有 26 個自由度(degree of freedom, DOF)的 3D 手部模型繪製出的影像計算而成的。輸入影像中的手與 3D 手部模型之間的差異會被公式化為目標函數,我們會使用粒子群最佳化(Particle Swarm Optimization, PSO)演算法去最小化這個目標函數,以求得一個具有 26 個維度的解。基於模型的方法並搭配 PSO 能為我們提供一個連續的解,也就是對於輸入影像序列能估測出連續的手部姿勢。
A model-based method that estimating the hand posture from a given monocular image sequence of a human hand is presented. The proposed method is markerless, that means it does not need special marker on hand. We compute image features, including silhouette and edge used in objective function, from a 26-DOF 3D geometric hand model and image sequence. The objective function is formulated to minimize the discrepancy between the input image and hand model, it is used to solve 26 dimensional parameters using Particle Swarm Optimization(PSO). Model-based method using PSO provides continuous solutions to our problem of estimating all articulations of hand posture.