基于深度学习的人体摔倒检测算法设计
Design of a human fall detection algorithm based on deep learning
李茁闻 1和晓军1
作者信息
- 1. 沈阳理工大学自动化与电气工程学院,辽宁沈阳 110159
- 折叠
摘要
针对传统检测算法对人体摔倒检测不准确以及实效性不够的前提下,提出一种改进的YOLOV7算法.将一种即插即用的Transformer模块插入到YOLOV7的骨干网络与检测头网络中.替换YOLOV7中高效聚合网络中的3*3卷积,在保证检测速度的同时提升模型的检测精度,其中平均精度提升12.32%、准确率提升5.01%、召回率提升3.33%.改进的YOLOV7人体摔倒检测模型可以满足不同应用场景中对各种设备的部署与高效地检测.
Abstract
An improved YOLOV7 algorithm is proposed to address the inaccurate and insufficient effectiveness of traditional de-tection algorithms in detecting human falls.Inserting Plug-and-play transformer module into YOLOV7 backbone network and detection head network.Replacing the 3*3 convolution in the efficient aggregation network in YOLOV7 improves the detection accuracy of the model while ensuring detection speed,with an average accuracy improvement of 12.32%,an accuracy improvement of 5.01%,and a re-call improvement of 3.33%.The improved YOLOV7 human fall detection model can meet the deployment and efficient detection of var-ious devices in different application scenarios.
关键词
计算机视觉/深度学习/YOLOV7/TransformerKey words
Computer vision/Deep learning/YOLOV7/Transformer引用本文复制引用
出版年
2024
国家科技期刊平台
NSTL
维普
万方数据