The Applied Research Of Hand Gesture Recognition In Human-Computer Interaction

Computers have been so tightly integrated with our everyday life that their new applications and hardware had been constantly introduced. However the man-machine interfaces mostly used at present are generally limited to keyboards, mice, light pen etc. Though familiar by people, these devices have their inherent limits in the speed and convenience with which we interact on computers. Recently, more and more interests are focus on human hand gesture recognition, which had shown its potential application in computer games, robot control and household appliances control etc, for man-machine interface. Furthermore the hand gesture technology had become an important research branch in the field of computer vision, pattern recognition and digital signal process etc.Our research work aims to supply an intuitionistic and comfortable keyboard-like input device for users to more easily and naturally manipulate a Computer Numerical Control system, e.g. CNC machine tools or Coordinate Measuring Machines. We employ hand gesture technology to achieve our target of the project. The key of this technology lies in the coding of interactive instructions input into computer with the flex statuses of 4 human fingers: index finger, middle finger, ring finger and small finger. Though aimed to be applied in a CNC system, we makes the 12 characters of a standard PC keyboard: "0～9", "." and "enter" as example to simulate our algorithm. The research work is decomposed into 3 parts: the pre-process of hand gesture image, feature extraction and recognition algorithm. Traditionally, palm centroid, statuses of fingertips and finger roots are the 3 mostly used features by preceding researchers in hand gesture recognition, with low correctness ratio. Different from their algorithms of feature extraction, this paper presents a new feature extraction algorithm for the design of classifier. This method firstly makes a straight line to fit the outer edge of little finger. Then the fingertips of the 4 fingers can be located through scanning along the direction perpendicular to the fitting line. The flex status of the fingers, determined by the relative fingertips position to that of middle finger, is figured out and used as the features for the succedent recognition algorithm.The traditional recognition algorithm employed in relative works of other researchers are generally based on the so-called matching technology, in which the extracted features are one by one matched with those in a pre-built feature library. In this paper Bayesian discriminant is employed as the criteria for the design of classifier.Our method is implemented under Matlab 6.5 with 60 hand gesture images, among which 85% are correctly recognized, indicating the effectiveness of this proposed method in this paper.