基于YOLOv5的手语手势识别系统  

Sign Language Gesture Recognition System Based on YOLOv5

作　　者：覃博铭 符文丝 唐梓航 QIN Boming;FU Wensi;TANG Zihang(Guilin University of Electronic Technology,Guilin 541004,China)

机构地区：[1]桂林电子科技大学,广西桂林541004

出　　处：《现代信息科技》2025年第6期121-125,共5页Modern Information Technology

摘　　要：目前,现有的聋哑人沟通技术存在识别准确度不高、设备烦琐等问题。文章提出了一种基于YOLOv5的手语手势识别系统。首先,对采集到的图像信息进行灰度化等预处理,利用OpenCV库的内置函数对图像进行局部块划分和特征提取,并对提取的样本进行RGB转换;其次,将经过预处理的图像输入到模型中,利用Adam优化器动态调整学习率,并采用K-means算法计算适合的锚点框值,从而提高模型的训练速度;最后,对模型输出的图像进行预测修正,通过非极大值抑制去除冗余边界框,保留最终的预测结果。实验表明,基于YOLOv5的训练模型在手语手势识别的准确度和识别率方面有较大提升。At present,the existing communication technologies for the deaf-mute individuals have problems such as low recognition accuracy and cumbersome equipment.This paper proposes a sign language gesture recognition system based on YOLOv5.Firstly,preprocessing such as grayscale conversion is carried out on the collected image information.The built-in functions of the OpenCV library are utilized to divide the images into local blocks and extract features,and RGB conversion is performed on the extracted samples.Secondly,the processed images are input into the model.The Adam optimizer is used to dynamically adjust the learning rate,and the K-means algorithm is adopted to calculate the appropriate anchor box values,thus improving the training speed of the model.Finally,prediction correction is carried out on the images output by the model.Redundant bounding boxes are removed through non-maximum suppression,and the final prediction results are retained.Experiments show that the training model based on YOLOv5 has a significant improvement in the accuracy and recognition rate of sign language gesture recognition.

关 键 词：YOLOv5 手势识别 OPENCV ADAM 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]