摘要
为实现对无人艇的自主回收,通过无人艇和悬浮托架的双向视觉导引,构建基于视觉和定位系统的无人艇自主对接系统。视觉对接系统通过检测网络分别检测出图像中的无人艇和悬浮托架。针对托架视觉检测的目标,利用位置定位信息排除非目标艇,采用单目标跟踪算法(CSR-DCF)对目标艇进行跟踪。为实现无人艇和托架的高精度稳定检测,在Tiny-YOLOv3的基础上,基于细节特征的需求,构建小尺度检测分支,提出一种定位增强轻量级检测网络,对于艇、托架的定位性能分别提升3.48%mIOU、4.32%mIOU,检测精度分别提升4.97%AP、6.17%AP。自主对接系统在二级海况下对接成功率达到90.9%。
To autonomously recycle USV,autonomous docking system for USV based on vision and position system with bidirectional visual guidance of USV and floating bracket was established.Vision-based docking system detected USV and floating bracket in images with detection network.According to the targets detected by bracket,position information was used to get rid of the interfering targets,the target USV was tracked with single target tracking algorithm(CSR-DCF).To realize precise and stable detection of USV and bracket,a location performance enhanced detection network was put forward.On the basis of Tiny-YOLOv3,the network added small scale detection branch for minutiae feature.Location performance for USV and brackets is increased by 3.48%mIOU and 4.32%mIOU while detection precision is increased by 4.97%AP and 6.17%AP.Success pro-bability of autonomous docking system reaches 90.9%under secondary sea state.
作者
徐海彬
刘畅
田建东
李小毛
XU Hai-bin;LIU Chang;TIAN Jian-dong;LI Xiao-mao(School of Mechatronic Engineering and Automation,Shanghai University,Shanghai 200444,China;State Key Laboratory of Robotics,Shenyang Institute of Automation,Chinese Academy of Sciences,Shenyang 110016,China)
出处
《计算机工程与设计》
北大核心
2021年第9期2606-2613,共8页
Computer Engineering and Design
基金
国家自然科学基金项目(91648118、61673254)
科技部重点研发计划基金项目(2018YFF0103400)。
关键词
自主对接
轻量级网络
目标检测
单目标跟踪
无人艇
autonomous docking
light-weight network
object detection
single target tracking
USV