基于深度学习的单目直线位移测量
Monocular linear displacement measurement with deep learning
石周 1李忠国 1车赛 1唐洪川 1高庙伟 1吴金坤1
作者信息
- 1. 江苏科技大学机械工程学院,江苏 镇江 212100
- 折叠
摘要
为进行基于机器视觉的位移测量研究,提出一种基于深度学习的直线位移测量方法.首先通过轻量化模型YOLOv5s-DSS用于实验小车的检测,然后通过Allometricl函数优化后的单目视觉定位模型用于小车定位,最后根据定位得到的起点和终点坐标,通过欧氏距离求解出目标运动的直线位移.结果表明,在3m内位移测量的相对误差在2%以内,证明了该方法可以有效实现运动目标的直线位移测量.
Abstract
A deep learning based linear displacement measurement method is proposed for machine vision based displacement measurement research.Firstly,the lightweight model YOLOv5s-DSS is used to detect the experimental car.Then,the monocular vision localization model optimized by Allometricl function is used to locate the car.Finally,the linear displacement of the target motion is solved by Euclidean distance according to the start and end point coordinates obtained from the localization.The results show that the relative error of displacement measurement within 3m is less than 2%,which verifies that the proposed method can effectively realize the linear displacement measurement of moving targets.
关键词
位移测量/目标检测/轻量化模型/目标定位/单目视觉Key words
displacement measurement/object detection/lightweight model/object localization/monocular vision引用本文复制引用
基金项目
江苏省产业前瞻与关键核心技术重点项目(BE2022062)
出版年
2023
NETL
NSTL
维普
万方数据