摘要
提出了一个新的水下声视觉图像预处理、分割和目标跟踪的处理系统框架。采用该系统框架，设计了一个基于前视声纳的智能水下机器人（AUV）声视觉目标探测跟踪系统，并描述了该系统的软、硬件体系结构。针对水下声视觉图像特点，分析了声纳图像的预处理方法，探讨了图像中特征信息的选取，构造了基于不变矩的仿射变换不变量，提出了基于组合特征的粒子权重分配方法，阐述了改进后的高斯粒子滤波（GPF）跟踪实现过程。海上实验验证了所提方法的有效性，证明所构建的探测跟踪系统具有较高的准确性和鲁棒性。
A new framework for pre-processing of underwater sonar data, segmenting of underwater sonar images and tracking of underwater moving objects was brought forward. Using the framework, a object tracking system with the sonar vision for autonomous underwater vehicles （AUV） was designed based on a forward looking sonar sensor, and its hardware structure and software system were described. The techniques for pre-process of sonar images were analyzed, the selection of the feature information in sonar images was investigated, and the affine transformation invari- ants based on invariant moments were constructed. The particle weight assignment method based on combination features was proposed, and the implementation of the improved Gaussian particle filter （GPF） tracking was expounded in detail. The object detection and tracing experiments were carried out. The results show that the system presented can be applied to underwater object detection and tracing, with the high real-time performance and accuracy.
出处
《高技术通讯》
CAS
CSCD
北大核心
2012年第5期502-509,共8页
Chinese High Technology Letters
基金
863计划（2008AA092301）,国家自然科学基金（51009040,E091002）和水下智能机器人技术国防科技重点实验室开放课题研究基金（2008002）资助项目.
