Research On Gesture Recognition Based On Vision

Human-Computer Interaction (HCI) based on visual has become an important topic of modern science and technology research. Because the gestures is diverse, multivocal, spatial, natural, intuitive and easy to understand, it has become a harmonious and natural HCI and one of the important means at the same time. The focus of the new generation of HCI has been from the static to the dynamic gesture recognition. And it has obtained favor of domestic and foreign researchers.This paper mainly studies the dynamic gesture recognition and its application in HCI. In the image preprocessing, the original image from the camera need to be binarized, dilated, eroded and filtered. Next, remove the background interference with skin segmentation method in the H channel of the HSV color space.In order to remove the redundant information to extract the gesture features precisely, we locate and remove the face area. Then, find the hand centroid that the important gesture characteristics by Camshift algorithm, and get the gesture feature sequence by quantizing and coding. Because of the low recognition rate with the traditional gesture recognition algorithm, this paper apply curve fitting of least-squares, and optimized it by segmentation to identify more gestures.According to the relative arbitrariness at the beginning and the end, as well as the characteristics of stable relatively in the mid-stage, this paper put forward an optimized scheme based on normal distribution as weighting factor aim to matching distance in the calculation process of DTW algorithm. The scheme can eliminate shake interference effectively during exercise of gestures to improve the recognition efficiency.Finally, this paper has set an experiment system of dynamic gesture recognition, and written programs based on OpenCV functions on the VC platform. Then the traditional DTW algorithm and the optimized DTW algorithm have been compared. The result shows that the optimized DTW algorithm has the better stability and the higher recognition rate. This topic could realizie that control word document with dynamic gestures under the real-time conditions, and complete a simple human-computer interaction.