Recognition Method For Gesture With Computer Vision

The technology of human-computer interaction is a hot field in recent years. There are mainly two research directions: speech recognition technology and body language recognition technology. Due to the theoretical content of body language identification is very broad,including image processing, computer vision, pattern recognition. Therefore, compared to the speech recognition, the area of body research is more challenging. Gesture interaction is the most innovative and reliable natural interaction machine undoubtedly, because the gesture is a object that constantly changing shape, so that research in this direction has become a very challenging research topic. This paper used computer vision method to achieve dynamic and static gesture recognition.First, this paper analyzed and compared the gesture segmentation algorithms. Due to the limitations of a single segmentation algorithm, this paper proposed through combination of color and region segmentation algorithm to segment the hand out from a complex background.Experimental results showed that this method can be very good split out gestures. After gesture segmented, we used DP algorithm to fit gesture with polygon fitting, then we completed the pretreatment of gesture. Because the geometric moment of recognition has a feature of high speed, so this paper extracted the features of static gestures by using geometric moments. Finally,we achieved the recognition of Arabic numerals 1 to 10 and completed the identification of static gestures by adopting the template matching algorithm. Then this paper analyzed and compared the particle filter with camshift tracking algorithm, for the particle filter algorithm will produce particle degradation, this paper adopted camshift tracking algorithm to track gestures. On the feature extraction module, due to the recognition rate of the geometric moments is very fast for the static gestures. This paper used geometric moments as static feature extraction. This paper used dynamic gestures trajectory tangent angle atdifferent time as the track feature. Taking into account the eight directions freeman algorithm has a small search range when searching, and traversing the template each pixel when each time search the edge elements, this way increased the amount of computation, so in order to achieve gesture feature extraction we proposed the improved Freeman algorithm to discrete gesture trajectory.In the end, this paper focused its attention on the improved HMM. Because the traditional HMM algorithm has three disadvantages, the improved HMM algorithm is used to train and recognize the gesture. Results show that this method of the improved Hidden Markov Model has a low complexity?high efficiency and accuracy of recognition, which also has a good practicability.