應用於穿戴式人機互動裝置之近距離即時手勢辨識技術

隨著科技的演進，人們與3C產品的互動方式逐漸由按鍵、觸控螢幕演進至體感操作，如Kinect、Xtion Pro等體感裝置，因為需要全身肢體控制，故其需要較長的距離才能使用。而近年穿戴式裝置的議題興起，如：Google Glass、Galaxy Gear等，將人機互動距離縮短至伸手可觸及的範圍以內，但目前此類裝置仍多為觸控操作，使用上較不直覺方便。因此，如何將「體感操作」結合至穿戴式裝置，讓其可透過直覺的手勢來操作，便成為ㄧ值得討論的問題。本論文提出ㄧ近距離手勢辨識演算法，主要針對頭戴式穿戴式裝置情境，使用深度影像資訊，實作出靜態、動態手勢辨識與手寫軌跡辨識。根據使用之情形，將15~50公分內的手部深度影像作為主要輸入來源，手部深度影像經過幾何學去雜訊及二值化後，使用HOG做邊緣特徵擷取，並將此邊緣特徵使用SVM分類以達到手勢辨識的效果，另一部分則使用指尖偵測演算法來獲取使用者欲執行操作之座標，並可將指尖經過之軌跡進行手寫數字辨識，達成基本操作的需求。本論文之方法未來可應用於穿戴式裝置，並可結合聯網電視或多種遊戲娛樂使用，提供一種新型態的互動方式。

關鍵字

體感操作；穿戴式裝置；手勢辨識；深度影像；手寫辨識； HOG ； SVM

並列摘要

With new technological advances, the human-machine control methods of the consumer electronic products are changed from conventional keyboards and touch screens to body actions. The depth-based motion sensors, such as Kinect, Xtion Pro., need sufficient distances between the users and the sensors with more than one meter, otherwise these sensors cannot appropriately work. In recent years, the wearable devices, such as Google Glass and Galaxy Gear, become more popular and thus the motion sensors of these devices need to detect users’ actions in near-eye distances. However, most of wearable devices are controlled by touch panels that are not intuitive and inconvenient. Therefore, how to combine gesture control and wearable devices to provide natural user interfaces will become an important issue for newly developed wearable devices, especially the glasses. To overcome the above-mentioned issues, this thesis proposes algorithms for near-distance hand gesture recognition. Our proposed techniques will focus on head-mounted wearable devices. Based on near-distance depth images, new gesture recognition techniques for wearable glass devices, including static gesture recognition, dynamic gesture recognition, and handwriting recognition modules, are developed. According to the use cases of wearable devices, the depth images are captured within 15 to 50 cm. We use histogram of gradients (HOGs) descriptors to represent features of hand gestures, then use the support vector machines (SVMs) to classify these features into meaningful gestures. Moreover, the proposed methods can detect the fingertips to allow users to type the numbers and perform handwritten recognition in the air. The proposed techniques can be applied for wearable human-computer control interfaces for the newly developed wearable devices, combined with interactive TV, games and interactive computer operations, to provide new experiences and applications.

並列關鍵字

motion control ； wearable device ； gesture recognition ； depth image ； handwriting recognition ； HOG ； SVM

參考文獻

[4] 賴彥成，基於深度影像與膚色之即時手勢辨識技術，碩士論文，國立台北科技大學資訊工程系，台北，2013

[6] Google Glass

[13] F. Gasparini, R. Schettini, “Skin segmentation using multiple thresholding,” Proceedins of the SPIE, Internet Imaging, 2006, pp. 1–8

[14] P. S. Hiremath, A. Danti, “Detection of multiple faces in an image using skin color information and lines-of-separability face model,” Int.J.Pattern Recognition Artif.Intell, 2006, pp. 39–61

[15] N. DALAL, B. TRIGGS, “Histograms of oriented gradients for human detection”. IEEE Computer Society Conference, Computer Vision and Pattern Recognition, vol.1, 2005, pp.886-893

國際替代計量

應用於穿戴式人機互動裝置之近距離即時手勢辨識技術

全文下載

主題瀏覽