基于改进的YOLOV4的手势识别算法及其应用

Gesture Recognition Algorithm and Application Based on Improved YOLOV4

针对基于特征提取的手势识别算法准确率低和速度慢的问题,提出了一种基于改进YOLOV4的手势交互算法.利用数据增强的方法解决了手势数据较少的问题,同时优化YOLOV4的网络模型,改变原特征图分辨率使其能够检测到更完整的手势特征,通过K-means算法重新计算先验框的尺寸以提高对不同大小手势的识别率.将该方法与原始的YOLOV4以及Faster R-CNN算法在不同的手势数据集上进行手势识别对比,并进行了手势交互实验.实验结果表明,基于YOLOV4的改进算法手势识别准确率更高,并且能识别复杂场景下的手势,检测速度可以达到32.3帧/s,能够满足实时要求.

Aiming at the problem of low accuracy and slow speed of gesture recognition algorithm based on feature extraction,a gesture interaction algorithm based on improved YOLOV4 was proposed.Use the method of data enhancement to solve the problem of less gesture data,at the same time,the network model of YOLOV4 was optimized,and the resolution of the original feature map was changed so that it can detect more complete gesture features,and the size of the prior frame was recalculated through the K-means algorithm to improve the recognition rate of gestures of different sizes.This method was compared with the original YOLOV4 and Faster R-CNN algorithms on different gesture datasets and gesture interaction was performed.The experimental results show that the improved algorithm based on YOLOV4 has higher accuracy in gesture recognition and can recognize gestures in complex scenes,the detection speed can reach 32.3 frame per second,which can meet real-time requirements.