Research On Dynamic Gesture Recognition Based On Leap Motion

In the last few decades, the mouse and keyboards help humans communicate with computers, and people have mastered and make good use of them. However, with the high speed development of multimedia technology, the traditional interactive tools can not satisfy humans’communicating needs anymore. People would rather use simple gestures, to make things effective, to communicate with computers. So, the study of the recognition of dynamic gesture is of important significance in present.This paper uses motion-sensing controller newly invented by Leap Co.--Leap Motion to collect information of dynamic gestures, including coordinate, acceleration and direction of fingers. Deal the collected information of moving dynamic gestures with operations like space diverting or normalizing, which normalize the information in space.Then, extract the features of those gestures. A dynamic gesture contains two features which represent spatial position and strokes including motion direction of gestures. Totally, we get 5200 features in the database which divided into 2 super-groups 26 groups after collected by leap motion and normalization.In this paper, dynamic gesture recognition is researched on two aspects. First, this article summarizes the data and results of dynamic gesture recognition experiments, which uses a single feature and bases on HMM model. Then the paper puts forward a model of dual channel based on the HMM model referencing the previous summary. According to the dynamic gesture recognition experiments in which 26 letters were defined as the dynamic gestures in this paper, the method can identify 98.4percent of the sample at average and 94.92percent of the testing sample...And the average time to identify each gesture is 0.18 second. Secondly, we use PM1K as the classifier to build a dynamic gesture recognition system and the RBF algorithm is implemented by hardware in this chip. It is faster to recognize than software as the chip has achieved truly parallel computing. The seven groups of dynamic hand gestures recognition results show that the system can identify a 120-dimensional feature vector gesture in 0.001171 second, and the recognition rate of the system can be as high as 96.57%.In general, this paper apply some new tools to studying dynamic gesture recognition both in software and hardware and enhance the recognition accuracy and speed, which has a certain value in gesture-based human-computer interaction applications.