基於軌跡辨識技術之人體姿勢自定與分辨研究

Abstract

鍵盤、滑鼠，是操作電腦不可或缺的設備，而隨著時代的進步，輸入設備不再侷限在此之上，如眼動儀的使用，運用眼球追蹤技術來控制滑鼠；語音輸入、辨識系統，能使較不熟悉鍵盤操作的使用者，能夠利用語音輸入設備達到打字的效果；觸控螢幕，讓手機、電腦的操作在手指滑動間即可達成，這些科技的發明，都讓電腦的操作更為人性化。而本研究係使用的微軟Kinect做為輸入端，讓使用者能自行輸入姿勢後再經本系統進行辨識，讓使用者以最直覺且習慣的方式操作電腦。本系統係以軌跡辨識為基礎，收集Kinect所提供的骨架資訊，再以決策樹的方式對使用者所輸入的姿勢進行儲存、分類與辨識，並在不造成使用者負擔的前提之下，以少量的事前訓練姿勢達到一定的辨識效果。Keyboard and mouse are the essential equipment to control computers. As time goes by, the way of controling computers is changing. For example , Eye tracker uses eye tracking to control mouse; Voice recognition let users use their own voice to input data as keyboard, providing a new way for those who not familiar with keyboard; touching screen, one of the most popular way to control computers and mobile phones, let users controlling their devices by just their fingertips. These technologies provide a more humanistic perspective way controlling computers. Kinect, the input of my research, using the whole body as a controller, with the corresponding special design system, this can reach a more intuitive way to control computer. In this research, We use Trajectory based approach and collecting skeleton data from Kinect to store and categorize user’s gestures. In this research, we construct a decision-tree-like structure to recognize users’ gesture, hoping using the less training data to get good preciseness.