基于视觉的手指语识别系统设计

Design of finger language recognition system based on vision

针对传统手指语识别系统识别速度慢、准确率低、运算量大等问题,提出一种基于MediaPipe框架与支持向量机(SVM)的轻量级手指语识别系统。该系统通过MediaPipe提取手指语图像关节特征点,使用支持向量机(SVM)中的径向基核函数(RBF)对提取的特征进行分类,以一对一的方法将基本的二分类SVM转变为多类SVM,从而实现手语识别目的。该文还用准确度、精度、召回率和F分数等指标对该系统的性能进行了评价。

Aiming at the problems of traditional finger language recognition system such as slow recognition speed,low accuracy and large amount of computation,a lightweight finger language recognition method based on MediaPipe framework and support vector machine(SVM)is proposed.The system extracts the joint feature points of finger language image through MediaPipe,classifies the extracted features using radial basis function(RBF)in support vector machine(SVM),and converts the basic two-class SVM into multi-class SVM by one-to-one method,thus realizing the purpose of sign language recognition.The proposed system performance is measured by accuracy,precision,recall rate and F score.