知识与财富的链接
针对传统手势识别方法计算量大、难以实时识别的问题,研究一种基于改进YOLO v5的轻量化手语检测识别方法.首先用Mobilenet v3-Small替换YOLO v5的主干网络;然后利用Ghost Conv模块和C3Ghost模块替换YOLO v5颈部网络中的Conv和Ghost模块;最后通过YOLO v5的预测部分生成预测框.在此基础上,利用k-means算法生成适合手势的先验框,加快网络检测手势.与其他网络算法对比分析可知,改进算法在保持精度基本不变的情况下可大幅减少网络参数,提高网络的检测速度.
A lightweight sign language detection and recognition method based on improved YOLO v5 are investigated to address the problems of excessive computation and difficulty in real-time recognition of traditional sign language recognition methods. Firstly, the backbone network of YOLO v5 is replaced by Mobilenetv3-small; then the conv and ghost modules in the neck network of YOLO v5 are replaced by Ghostconv module and C3Ghost module; finally, the prediction frame is generated by the prediction part of YOLO v5. Based on this, the kmeans algorithm is used to generate a priori frame suitable for gestures to speed up the network to detect gestures. A comparative analysis with other networks shows that the improved algorithm significantly reduces the network parameters and increases the prediction speed of the network while keeping the accuracy largely unchanged.