改进Bot-SORT的边坡落石监测方法
Improved Bot-SORT for Slope Rockfall Monitoring
王晓青 1阎吉 1张德育1
作者信息
- 1. 沈阳理工大学信息科学与工程学院,沈阳 110159
- 折叠
摘要
针对边坡落石监测中存在的目标尺寸小、石块与背景特征差距小、落石目标运动速度快等问题,提出一种基于检测的改进Bot-SORT多 目标跟踪算法.在检测部分对YOLOv7模型进行改进,引入注意力机制,提升模型对石块特征的提取能力,并使用归一化高斯Wasser-stein距离作为真值框与预测框的距离度量方式,降低模型对小目标的漏检率;在跟踪部分引入GIoU距离匹配方式,有效跟踪快速运动的落石.通过实景拍摄及Unity仿真方式建立训练及测试数据集,消融实验和对比实验结果表明,本文改进算法能够有效提高落石的检测率和跟踪精度.
Abstract
A detection based improved Bot-SORT multi-objective tracking algorithm is proposed to address the issues of small target size,small difference between rock and background features,and fast movement speed of rockfall targets in slope rockfall monitoring.In the detection section,the YOLOv7 model is improved by introducing attention mechanism to enhance the model's ability to extract stone features.Normalized Gaussian Wasserstein distance is used as a metric for the ground truth box and target bounding box to reduce the model's missed detection rate for small targets.The GIoU distance matching method is introduced in the tracking section to effectively track fast falling rocks.The training and testing dataset is established through live shooting and Unity simula-tion.The results of ablation experiments and comparative experiments show that the improved algo-rithm in this paper can effectively improve the detection rate and tracking accuracy of falling rocks.
关键词
多目标跟踪/落石监测/YOLOv7/Bot-SORTKey words
multi-object tracking/rockfall monitoring/YOLOv7/Bot-SORT引用本文复制引用
基金项目
辽宁省科学技术基金项目(2022JH1/10800085)
出版年
2024
NETL
NSTL
万方数据