隨著人臉偵測技術的快速發展與應用越來廣泛，配合智慧型手機日益達 與其運算速 度也有飛快的進步，因此即時人臉偵測使用在智慧型手機上應越來 越多，本研究使用 Android系統的智慧型手機來進行即時人臉偵 測並且希望除了測人臉功能以外，還可準確地量轉動的實際角度並使用偵出來資訊來執行一些動作。本研究透過 Google Vision API提供的服務來進行優化，使用 提供的服務來進行優化，使用 Google API提供的人臉轉動 提供的人臉轉動 Euler值來訓練回歸方程式並且分析本實驗室親自建設 值來訓練回歸方程式並且分析本實驗室親自建設 的人臉轉動資料庫與網路上之優劣。由於智慧型手機在偵測大角度(大於 60 度)偵測時並非十分穩定，因此本研究也對問 題做出優化的動作。最 後本系統為了讓使用者有更好的體驗，在整個中也詳細考慮各種狀況與設計使用者在操作上可以更快速簡單。

With the rapid development and application of face detection technology, the adoption of smart phones is getting more and more advanced , and the speed of computing is also increasing rapidly. This real-time face detection use on smart phones more and more applications, this study uses Android’s smartphone for real-time face detection and hope in addition to the function to detect face, but also can accurately measure the actual angle of the face rotation and use the detected angle information to perform some actions. This study optimizes the services provided by the Google Vision API, uses the face rotation Euler value provided by the Google API to train the regression equation and analyzes the advantages and disadvantages of our laboratory built the face rotation database and the face rotation database on the network. Since smartphones are not enough stable when detection large angles (greater than 60 degrees) , this study also optimizes the problem. Finally, in order to give the user a better user experience, in this system the user’s various conditions are also considered in detail, and design user can operate faster and easier.

致謝--------------------------------------------------------------i
中文摘要---------------------------------------------------------ii
ABSTRACT-------------------------------------------------------iii
目錄-------------------------------------------------------------iv
圖目錄-----------------------------------------------------------vi
表目錄---------------------------------------------------------viii
第1章 緒論-----------------------------------------------------1
1.1 研究動機---------------------------------------------------1
1.2 研究方法---------------------------------------------------3
1.3 論文架構---------------------------------------------------6
第2章 研究背景-------------------------------------------------7
2.1 Android作業系統介紹----------------------------------------7
2.1.1 Android版本----------------------------------------------7
2.1.2 應用程式介面(Application Programming Interface ,API)-----8
2.2 人臉偵測 (Face Detection)----------------------------------9
2.2.1 灰階處理 (Gray Level Processing)------------------------10
2.2.2 直方圖等化 (Histogram Equalization)---------------------10
2.2.3 人臉偵測方法--------------------------------------------13
2.3 Google Vision API----------------------------------------17
2.3.1 Google Vision API簡介-----------------------------------17
2.3.2 人臉偵測API---------------------------------------------17
2.3.2.1 臉部方向 (Face Orientation)---------------------------17
2.3.2.2 Landmark---------------------------------------------18
2.3.2.3 臉部表情----------------------------------------------19
2.4 USB Port--------------------------------------------- ---20
2.4.1 USB OTG(On-The-Go)----------------------------- -------20
2.4.2 USB Video Class (UVC)----------------------------------21
第3章 系統設計------------------------------------------------23
3.1 Euler分析-------------------------------------------------23
3.1.1 回歸方程式----------------------------------------------23
3.1.2 資料庫品質優劣分析---------------------------------------26
3.2 相異手機分析----------------------------------------------28
3.3 系統架構--------------------------------------------------32
3.3.1 單鏡頭模式----------------------------------------------33
3.3.1.1 單鏡頭水平角度偵測-------------------------------------35
3.3.1.2 單鏡頭垂直偵測-----------------------------------------36
3.3.2 雙鏡頭模式----------------------------------------------39
3.3.2.1 初始化子系統------------------------------------------39
3.3.2.2 雙鏡頭模式水平角度偵測---------------------------------42
3.3.2.3 雙鏡頭模式結束偵測子系統--------------------------------44
3.3.2.4 雙鏡頭垂直偵測-----------------------------------------45
第4章 成果展示------------------------------------------------47
4.1 實驗需求-------------------------------------------------47
4.2 Android應用程式執行結果-----------------------------------48
4.2.1 單鏡頭模式執行結果---------------------------------------48
4.2.2 雙鏡頭模式執行結果---------------------------------------50
4.2.2.1 未連接UVC裝置-----------------------------------------50
4.2.2.2 已連接UVC裝置-----------------------------------------50
4.3 數據比較--------------------------------------------------54
第5章 結論與未來展望-------------------------------------------56
5.1 結論------------------------------------------------------56
5.2 未來展望--------------------------------------------------57
參考資料---------------------------------------------------------58

[1] Kimio Hamasaki , Toshiyuki Nishiguchi , Reiko Okumura , Yasushige Nakayama and Akio Ando , “A 22.2 multichannel sound system for ultrahigh-definition TV (UHDTV)”, NHK Science & Technical Research Laboratories, Tokyo, 157-8510, Japan, April 2008
[2] 正常的雙眼視覺 (資料出處: http://www.wunching.com.-tw/img/Books_files/B390-9789862369258-trial.pdf)
[3] Android , Wikipedia (資料出處:https://zh.wiki-pedia.org/wiki/Android)
[4] Android歷史 (資料出處:https://www.android.com/intl/zhTW_tw/history/#/-marshmallow)
[5] 灰階處理(Gray Level Processing) (資料出處: https://cg2010studio.com/2011/06-/06/opencv%E8%BD%89%E6%8F%9B%E5%BD%B1%E5%83%8F%E7%82%BA%E7%81%B0%E9%9A%8E-transform-image-to-gray-level/comment-page-2/)
[6] Wu Zhihong and Xiao Xiaohong , “|Study on Histogram Equalization” , 2011 International Symposium on Intelligence Information Processing and Trusted Computing
[7] 積分影像(Integral Image) (資料出處: https://cg2010studio.com/2012-/04/24/%E7%A9%8D%E5%88%86%E5%BD%B1%E5%83%8F-integral-image/)
[8] Paul Viola and Michael Jones “Rapid Object Detection using a Boosted Cascade of Simple Features” , 2001
[9] Google API for Android (資料出處:https://-developers.google.com/android/)
[10] Google Vision API (資料出處: https://-developers.google.com/vision/)
[11] Google Vision API for Face Detection (資料出處: https://developers.-google.com/vision/face-detection-concepts)
[12] USB Implementers Forum , “On-The-Go Supplement to the USB 2.0 Specification” December 5,2006
[13] USB Implementers Forum , “UVC 1.5 Class specification” , August 9 , 2012
[14] Databases for Face Detection and Pose Estimation (資料出處:http://robotics.-csie.ncku.-edu.tw/Databases/FaceDetect_PoseEstimate.htm)