專利商品網


    快速搜尋        


zoomin      

技術名稱 Technology
發明人 Inventor
黃百瑞, 林昌鴻, 李佳翰,
所有權人 Asignee 國立臺灣科技大學

專利國家 Country 申請號 Application No. 專利號 Patent No. 中心案號 Serial No.
中華民國 102136408 I499966 1020050TW0
美國 14/133,650 US9256324B2 1020050US0
  點閱數:862

技術摘要:
With the development of electronics in recent years, people started to concern with the issue of quality and capability that electronic devices can provide to users. One of these issues is that electronic devices can provide a human machine interface (HMI) that helps the users to interact friendly with these devices.
In this application, we proposed a novel method that can correctly detect the user’s fingertips when the user is in different complex scenarios. The proposed system is composed of seven components. In the first component, we used background subtraction to obtain the probable human body objects in the scene. Second, the human faces are detected by using the AdaBoost learning algorithm with Haar-like features. Once the human face is detected, in order to provide better performance in terms of the computation time, we would use Camshift tracking to continuously obtain the position of the user’s face. Then, the position of the face would be used to render the virtual buttons on the screen through augmented reality, and its depth value is used to automatically update the threshold value for the background subtraction.
For the fourth component, in order to accurately acquire the fingertip regions, we would first use a body mask to cover the human face and human body. Then, we detect the skin color regions in the foreground objects. However, whether the edges of skin color regions is smooth or not will acutely affect our system. Therefore, we used the Gaussian filter method and morphological image processing techniques to obtain complete skin color regions of the human body objects. After obtaining the hand areas, we used the fingertip properties which have defined in accordance with our observation to correctly detect the fingertips of both hands, and filter out the non-fingertip regions. Then, the tracking procedure is executed to analyze the motion of each fingertip. The final component, "Click" event recognition, is to check whether if the user clicks the virtual button or not according to our defined criterions.


相關圖片: