Research On Fingertip Detection Algorithm In The Process Of Motion

The research on the problem of fingertip detection is the key to the accurate execute of the user in the interface.Fingertip detection involves many aspects of image processing,including image preprocessing and fingertip detection.Due to complexity of environment and the change of hand shape,accurate fingertip detection is still challenging with the restriction of real-time.Existing methods mainly are skin color segmentation or motion segmentation to extract the hands in the foreground extraction.The threshold of the former is not easy to set and has many interference factors.The latter is poor in adaptability.Scholars have made innovations based on common methods and achieved certain results.For example,hybrid Gaussian models based on YCb Cr space is more and more effective but ignore the connection between point of neighborhood.Simply considering the change in the point of time causes the target edge is jagged when changing environment.In this paper,the neighborhood pixel connection weights are integrated into the adaptive mixture Gaussian model to obtain the motion area.Mixed Gauss skin color model is improved by adding the spatial neighborhood information of the object pixels.The Gauss component of adjacent pixels is obtained by the maximum posterior probability of the class that makes time and space unify so as to make up the shortcomings of the original algorithm.Reduces the number of Gauss distribution adaptively to make it meet the real-time performance.A shadow removal algorithm with a weighted color calculation model is designed to reduce the weight of the color components and brightness interference.Comparison experiments show that the improved foreground split F-value has a 0.05-0.27 increase.Existing monocular fingertip detection algorithms have more or less interference points and artificially set the threshold problem.The threshold value cannot be adjusted automatically during the process.Some scholars have tried to eliminate interference with a combination of algorithms but the effect is not obvious enough.This article uses the improved center-of-gravity distance method to detect the position of the fingertip to remove the interference point.From the overall characteristics of the hand that the interference point is eliminated according to the circular arc nature of the fingertip point.Calculate the average error of the fingertip point and its corresponding arc.The smaller the error is the more likely it is to be a fingertip.Experiments with different database from the foreground segmentation show that the average accuracy is 97.26% of fingertip detection in the improved method with complex background and fingertip deformation.The average processing time of per frame is 23.43 ms.The fingertip filtering is implemented efficiently and quickly,which shows strong robustness and meets the real-time requirements of dynamic detection.Finally,the detected fingertips are used in gesture recognition as a simple application.