| 摘要: | 在這個時代，科學技術的進步比人們想像的要快得多，人與機器
之間的交互不限於鍵盤鼠標和屏幕之間的交互。過去，我們認為只有
電影才能看到的虛擬介面和或是聲光線效果並不用再依賴演員的想
像力，相反地，現今的科技可以通過虛擬現實的技術為用戶提供身臨
其境的體驗。諸如虛擬實境頭盔或是手套等不同的攝像頭和傳感器為
用戶帶來了全新的體驗。科學技術的進步來自人們的需求，在當今社
會， 娛樂已成為生活中不可缺少的一部分。
Kinect 是由微軟開發的深度相機， 可以通過捕捉深度信息來追踪人
體的骨骼。通過識別人體關節的位置，用戶可以利用肢體的動作與機
器進行交互， 而我們利用並通過Kinect 辨識深度資訊的功能實現了
虛擬打擊樂器的設計。
由於硬件限制， Kinect 相機的分類器無法確定蒙面的肢體或微妙而
快速的移動。例如，在演奏打擊樂器的過程中，我們經常使用手指的
細微動作來改變敲擊的位置，然而攝影機卻無法捕捉如此細微的動作，
因此，我們希望通過在木槌上安裝六軸慣性傳感器以獲取加速度以及
角加速度的資料來解決這些問題。
近年來機器學習的迅速發展使其無法忽視它，機器在不同的領域裡都
取得了相當的成就。在本文中，我們嘗試通過機器學習方法來處理複
雜手勢識別的任務。
我們將藉由設計虛擬樂器來收集手部運動的三軸加速度和角加速度,
並以這些數據將做為手勢辨識的依據。當用戶執行虛擬樂器演奏時，
用戶可以通過在空中寫入來輸入手勢，並通過識別結果執行命令以改
變音調或音調。我們認為，多軸信號與機器學習的結合除了可以彌補
攝影機天生的缺陷之外,也擴大了人們對於人機交互的想像。;In this era, the progress of science and technology is much faster than people think. The interaction between humans and machines is not limited to the interaction between the keyboard mouse and the screen.
In the past, we believe virtual interfaces and acoustic lighting effects that can be only seen in the movie will no longer depend on the imagination of actors.
Instead, technologies nowadays can provide users with immersive experiences through virtual reality technology.
Different cameras and sensors like virtual reality headset or glove bring users a whole new experience.
The advancement of science and technology comes from people′s needs.
In today′s society, entertainment has become an indispensable part of life.
Kinect is a depth camera developed by Microsoft that can track the body′s skeleton by capturing depth information.
By recognizing the position of human joints, users can use the movement of the limbs to interact with the machine, and we have realized the design of virtual percussion instruments through these depth information.
Due to hardware limitations, the Kinect camera′s classifier cannot determine the masked limbs or the subtle and fast movements.
For example, in the process of playing percussion instrument, we often use the subtle movements of our fingers that can not be captured by camera to change the position of tapping .
Therefore, we hope to solve these problems by installing six-axis inertial sensors on mallet to obtain acceleration and angular acceleration data
The rapid development of machine learning in recent years has made it impossible to ignore, and it has made considerable achievements in different fields.
In our thesis, we will collect triaxial accelerations and angular accelerations of hand movements by designing virtual instruments and use these signals as the basis dataset for gesture recognition.
When the user play a virtual musical instrument, the user can input a gesture by writing in the air, and execute a command such as changing the pitch or tone according to the recognition result.
In our opinion, the combination of multi-axis signal and machine learning not only compensates for the inborn defects of the camera, but also expands people′s imagination for human-computer interaction. |
