控制環境下的手形辨識

Abstract

在這篇論文裡，針對連續動作的正面手部影像序列，我們嘗試建構一套手形辨識系統，以自動化方式分析二維手部影像並進而估測三維手形狀態。其中，我們利用影像處理技術從五指張開手部影像中萃取出手部特徵模型，並利用手部特徵模型追蹤後來影像中每一根手指所相對應的區域，來加以分析每一根手指所對應的特徵資訊。之後，根據空間域上擷取出的二維特徵資訊以及時間域上手形變化的連續性，我們利用貝氏決策定理來估測最佳的三維手形狀態。

In this thesis, we try to construct a hand pose recognition system, which can analyze a sequence of 2D images and reconstruct the corresponding 3D hand poses automatically. In the analyzing stage, we apply image processing techniques to build a hand feature model based on the first image frame, in which the hand pose is always in an “open” status. Based on this hand feature model, we detect finger regions and finger features in the following images. Finally, we use Bayesian decision theory to estimate the status of the 3D hand pose, by combining the 2D features in the spatial domain and the hand motion in the temporal domain.