一种轻量化的排球自垒姿态检测算法
A lightweight posture detection algorithm for volleyball self-digging
凌驯 1陶青川1
作者信息
- 1. 四川大学电子信息学院,成都 610065
- 折叠
摘要
边缘设备有限的存储和处理能力,在实际应用中难以部署较为复杂的YOLOv7pose姿态检测模型.对YOLOv7pose进行了一系列轻量化处理,使用FasterNet的主干网络重构YOLOv7pose的特征提取网络,将特征提取后的输出应用CBAM注意力机制来弥补精度上的损失,最后对冗余的多尺度检测头进行删减,实验表明改进后的轻量化网络较原网络的参数量下降了2/3,计算速度提升了2.5倍,精度仅减少了3.8%,能够满足边缘设备实时检测排球对墙自垒过程中的人体姿态情况.
Abstract
Due to the limited storage and processing capabilities of edge devices,it is difficult to deploy a more complex YO-LOv7pose pose detection model in practical applications.The article conducted a series of lightweight processing on YOLOv7pose,using the backbone network of FasterNet to reconstruct the feature extraction network of YOLOv7pose.The output of the feature ex-traction was applied with a CBAM module to compensate for the loss of accuracy.Finally,redundant multi-scale detection heads were deleted.Experiments have shown that the improved lightweight network reduces the parameter quantity by 2/3 compared to the original network,increases the computing speed by 2.5 times,and reduces the accuracy by only 3.8%.It can meet the real-time detection of human posture during the process of volleyball self-digging by edge devices.
关键词
姿态估计/轻量化/边缘设备/YOLOv7poseKey words
pose estimation/lightweight/edge devices/YOLOv7pose引用本文复制引用
出版年
2024
NETL
NSTL
万方数据