Research On Static Gesture Recognition Based On Visual And Geometric Features

The hand gesture is a human beings'natural language which has the characteristics of nature,simplicity,convenience and high real-time.It is not only unrestricted by region,environment and culture,but also more convenient than language communication to some extent.Gesture recognition technology emerge as the times require.Therefore,many researchers in the field of human-computer interaction devote themselves to judging the message by hand gesture.Because of the flexibility and dynamics of gesture morphology,based-vision gesture recognition has become the mainstream trend in practical application.In this paper,the based-vision gesture recognition method is deeply studied under complex background.In this paper,a series of researches and experiments are carried out on the key technologies of gesture segmentation,feature extraction and recognition matching.In the phase of gesture segmentation,an improved gesture segmentation algorithm is proposed to solve the problem that the image is not well segmented due to the influence of noise.Firstly,skin color model is built to get threshold.Then,the image is filtered locally according to the neighborhood information of the pixel.In order to get better segmentation effect,aiming at the defect of two-dimensional Otsu algorithm when choosing threshold,combining skin color threshold to optimize threshold,and the best threshold is calculated by adding the sum of squares between groups as fitness function.The experimental results show that the improved segmentation algorithm improves the segmentation effect of gesture image.In the feature extraction stage,aiming at the problem that low recognition rate caused by the influence of scaling and rotation of gesture,a HGD feature extraction method based on the geometric distribution of gesture is proposed.Firstly,standardize the segmented gesture image.Secondly,find the gesture main direction and calculate width-to-length ratio of gesture.The similarity function is used for preliminary recognition.Then,the coordinates of the contour points of the gesture in the Cartesian coordinate system are counted,and the final gesture is identified by using the modified Hausdorff distance as a similarity measure.In order to prove that the hand gesture features extracted by HGD algorithm have better performance,a series of experiments are designed for feature matching and recognition.Firstly,the performance of the algorithm is analyzed experimentally.The experiments show that HGD algorithm is not affected by illumination intensity and gesture background and has better recognition rate.In addition,the recognition experiments are compared with three similar feature extraction algorithms,The experimental results show that this method recognize gestures in various situations quickly and accurately.Under certain conditions,the average recognition rate is 92.89%,the average error recognition rate is 3.53%,and the recognition speed is 4.2 times faster than the similar algorithms.