利用雷射光斑成像裝置建構姿勢辨識系統

Abstract

姿勢辨識是電腦智慧中最重要的一環，能夠使電腦對於人類的動作姿勢進行判讀，幫助人類與機器之間的互動，現今辨識領域有許多的辨識演算法，但大多數皆過於複雜和費時，所以本論文的研究方向是利用動態時間扭曲演算法(Dynamic Time Warping ,DTW)的實現，來建構一個高準確性的姿勢辨識系統。 本論文主要研究為利用雷射光斑成像裝置Kinect當作偵測器，追蹤人體骨架以及紀錄人體關節座標，將輸入資訊與先前紀錄起來之關節座標的參考資訊比對，藉由DTW演算法的比對，能夠計算兩組資訊之間的相似度，並由高相似度的判讀達成高準確性辨識，以實現利用雷射光斑成像裝置建構姿勢辨識系統。

Gesture recognition is one of the most promising fields in computer science, and touch-input technology. It’s applications to sport gaming on a TV set such as Microsoft Kinect is well known. It also facilitates the interactions between computer and human beings. Most of algorithms for gesture recognition are too complicated and time-consuming. In this thesis, we use dynamic time warping algorithm to compute data that are less time-consuming and better accuracy. We use Kinect, a tool for laser-speckle imagery, containing a light source and a detector to obtain data for tracking skeleton and joint. Comparing these data with the reference data by DTW algorithm, we can figure out the similarity and the difference between two groups of data. Hence, our methods of gesture recognition are faster and better accuracy.