由於科技日新月異的進步,對於電腦或是電子機械的操控已不再局限於傳統的鍵盤、滑鼠、搖桿等,較不直覺化的操作,現今利用手勢作為直覺化的操作方式已有越來越受歡迎的趨勢。其應用有機器人控制、家電控制與互動式遊戲等多方面,使得生活更加便利。 手勢作法可分為基於影像與非影像等兩類。於非影像的方式中,使用者須穿戴具有感測器的數據手套,其感測器會對手部活動範圍造成影響,讓使用者在使用上較為缺乏自然性,而不易使用;感測器電量之持久性、成本較高等問題使得數據手套發展受限。如果是使用影像的方式,則可免除上述之問題。影像方面又可分為數位影像與深度影像,而數位影像與深度影像相較之下,對於低光源變化、多重膚色區域判別以及前後背景分割問題,使得數位影像處理方式,較不易識別手部區域。 因此本論文提出了基於深度影像的方法來識別使用者手勢,其過程包含手部分割、特徵提取、模糊推論(Fuzzy Inference)來識別手勢,達到直覺化操作的目的。而由實驗結果顯示本論文所提之方法可以識別出美國手語(American Sign Language, ASL)的數字手勢0到9,此方法平均的識別準確率達94.4%,具有一定的水準。
Due to the rapid progress of science and technology, computer or electronic mechanical manipulation is no longer confined to the traditional keyboard, mouse and joystick etc., which are not intuitive. Nowadays, using gestures as an intuitive mode of operation has been increasingly more popular. Its applications include robot control, appliance control, interactive games and other aspects, making life more convenient. Gestures practice can be divided into two categories based on images and non-images. In the non-images mode, the user should wear gloves which can sensor data, but the sensor will affect the motion range of hands so to make users lack naturalness of use and not easy to use. Persistent power of the sensors, high cost and other issues make the data glove development limited. If video images are used, you can relieve the problem described above. Images aspect can be divided into digital images and depth images; when digital images are compared with depth images, for low light changes, multi-color region discrimination, as well as before and after the background segmentation issues, the processing mode of digital images is more difficult to identify the hand area. Therefore, this paper presents a method based on the depth image to identify the user gesture, and the process comprises the hand segmentation, feature extraction, fuzzy inference to recognize gestures and achieve the intuitive operation. And the experimental results show that the method proposed by this paper can recognize digital gestures 0-9 of American Sign Language (ASL), so average recognition accuracy rate by this method is 94.4%, achieving a certain standard.