一個用於虛擬鍵盤之手勢識別框架;A Gesture Recognition Framework for Virtual Keyboard Applications

NCUIR > College of Electrical Engineering & Computer Science > Graduate Institute of Computer Science and Information Engineering > Electronic Thesis & Dissertation > Item 987654321/74638

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/74638

Title:	一個用於虛擬鍵盤之手勢識別框架;A Gesture Recognition Framework for Virtual Keyboard Applications
Authors:	古安徒;Enkhtogtokh, Togootogtokh
Contributors:	資訊工程學系
Keywords:	SMART TV虛擬鍵盤;Gesture event detection;Gesture event recognition;SMART TV Keyboard
Date:	2017-07-14
Issue Date:	2017-10-27 14:34:33 (UTC+8)
Publisher:	國立中央大學
Abstract:	了解多媒體內容在不同環境下的使用，特別是在日常生活中實現智能和簡單的使用是目前最有用的研究課題之一。我們提出了一種有效的機器學習方法來分析多媒體內容處理手勢事件的檢測和辨識。我們的機器學習方法具有基於經過深入研究的技術的訓練可靠的機制，例如Procrustes分析，局部和全局表徵的組合，線性形狀模型，並且應用於SMART TV虛擬鍵盤。在手勢事件檢測中，先定義點集的空間配置對於幾何是重要的。通常，增加這些點具有線性增量計算複雜度的算法的穩建性提高。我們考慮局部變形的組成的參數化，其解釋了實體之間的形狀和全局變換之間的差異，這說明了特定形狀作為手指和手的整體放置。在這項研究中，我們處理手勢事件檢測特別是指尖手勢檢測，以獲得智能和先進的技術使用。我們最新的視覺鍵盤可以是目前 SMART TV遠程控制的取代品。它更便宜，因為我們不需要實體裝置像傳統鍵盤、遙控器等，並且也不需要提供能源，例如電池等。深層結構學習（深度學習）技術的興起近年來受到人工智能（AI）的關注。自動編碼技術（Autoencoder）是有希望的方法之一，尤其是深層結構化架構，通常用於縮小尺寸。例如，無人駕駛車就是最好的使用深度學習的例子。當然，我們的人類視覺在我們的感知系統上也有類似的過程。在這裡我們提出了一般技術。特別地，在我們提出的神經科學動機的方法中，考慮了非平坦（彎曲）和嘈雜的解決非線性機器學習方法。從神經科學到了今天，我們對人類的看法有著重要的成果，我們的眼睛是如何工作的，特別是在通量條件下才能實現現實世界。在神經科學中，發現當我們的眼睛將信號傳遞給大腦時，大腦具有作為尺寸縮小處理的特殊過程。未來，機器將在觸摸，視覺，聽覺，味覺和氣味方面取得進步。計算機視覺的最終目標是使計算機具有人類眼睛和大腦的能力 - 甚至以某種方式超越和協助人類。它引導我們系統地研究微分幾何，使差分歧管理論。詳細地說，為了實現這一點，首先我們的數學模型必須將整體流形學習視為彎曲（非平坦）歧管案例，而現實世界數據則是雜訊處理模型受到最多關注。我們提出了融合數學模型，解決了曲率數據，異常值檢測，成本優化，自動參數選擇和样本外擴展，將其應用於機器。異常值檢測的實際應用範圍廣泛，如機械故障，系統行為，人為錯誤，人口自然偏差，大數據，高維數據作為深度點雲和非線性（多維）學習等。我們主要集中在非線性線性學習應用程序作為尺寸減少由於自然要求識別雜訊，以保留有意義的主要數據。沒有這種技術，尺寸減小方法是不可能獲得雜訊數據的正確結果。在黎曼空間的自然結構中，已經被認為是高維數據，這意味著在更實際真實的情況下工作，例如3D點雲數據具有非常重要的開始機制。 ;Understanding multimedia content uses on different context specially to achieve smart and easy usages in daily life is one of the most useful researching topics currently. We have proposed an effective machine learning method to analyze multimedia content addressing gesture event detection and recognition. Our machine learning method has training robust mechanisms as based on well-studied techniques such that Procrustes Analysis, Combination of Local and Global Representations, Linear Shape Model, and application to SMART TV Virtual Keyboard. The spatial configuration of a predefined set of points is important for Geometry in gesture event detection. Generally, increasing these points have improved robustness of algorithm with a linear incremental computational complexity. We consider the parameterization of composition of local deformation that accounts for the differences between shape across identities and global transformation that accounts for the overall placement of particular shape as fingers and hands. In this research, we address gesture event detection specially fingertip gesture detection to get smart and advanced usage of technology. Our modern vision keyboard could be a good next generation replacement of SMART TV remote control. It can be more economical as we don’t need physical object like traditional keyboard, remote control and their energy resources like batteries. Rise of deep structured learning (deep learning) technique is taking much attention in recent years for artificial intelligence (AI). Auto encoding technique (Autoencoder) is the one of the promising approach especially with deep structured architecture, typically for the purpose of dimensional reduction. As example driverless car is the best usage of it. Naturally, our human vision has the similar process on our perception system. Here we proposed the general technique for it. Specially, non-flat (curved) and noisy tackling non-linear machine learning method is taken into account in our proposed method with Neuroscience motivations. From Neuroscience, today, we have the important results about human perception that is how do our eyes work with brain especially under flux conditions to perciept the real world. In Neuroscience, it is discovered as when our eyes transfer the signal to brain, the brain has the special process that is the dimensional reduction processing. In the future, machine will make advances in touch, sight, hearing, taste, and smell. The ultimate goal of Computer Vision is for computers to have capability of human eyes and brains-or even to surpass and assist the human in certain ways. It leads us to systematically study the differential geometry such that differential manifold theory. In detail, to accomplish it, first our mathematical model must consider the general manifold learning as curved (non-flat) manifold case and real world data as noise handling model are taken the most attention. We have proposed the fusion mathematic model which solved the curvature data, outlier detection, cost optimization, automatic parameter choosing, and out-of-sample extension to apply it for machine. Practical application of outlier detection is widely ranged as mechanical faults, system behavior, human error, natural deviations in populations, big data, and high dimensional data as depth point clouds and non-linear (manifold) learning, etc. We mainly focused on non-linear learning application as dimensional reduction due to natural requirement to recognize noise in order to preserve the meaningful main data. Without this technique, dimensional reduction approaches is impossible to gain correct result on noise data. In the natural structure of Riemannian space, it is already considered the high dimensional data, which means working in more practical case, such as 3D point cloud data that is very important to have outset mechanism.
Appears in Collections:	[Graduate Institute of Computer Science and Information Engineering] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	183	View/Open

社群 sharing

Loading...