SIA OpenIR  > 光电信息技术研究室
基于移动终端的捷联成像末制导技术研究
Alternative TitleStrap-down Imaging Terminal Guidance Technology Based on Mobile Device
徐峥
Department光电信息技术研究室
Thesis Advisor罗海波
Keyword目标跟踪 深度学习 轮廓检测 制导解算 移动终端
Pages104页
Degree Discipline模式识别与智能系统
Degree Name博士
2020-05-19
Degree Grantor中国科学院沈阳自动化研究所
Place of Conferral沈阳
Abstract本文以复杂背景下运动小目标的跟踪和制导为应用场景,开展了基于移动智能终端的捷联成像末制导技术研究,利用移动智能终端的低成本传感器和处理器部件,完成捷联成像制导组件的设计开发与技术验证。主要研究工作总结如下:(1) 围绕复杂背景下运动小目标跟踪问题,研究了目标模板选取和跟踪稳定性之间的关系,在此基础上,为了提高对复杂背景下运动小目标跟踪的稳定性,提出了一种基于深度学习的目标模板提取方法。首先设计了深度轮廓检测网络用来提取目标轮廓,多组实验证明了本文设计的深度轮廓检测网络提取到的轮廓信息优于其他方法。然后,选取目标轮廓围成的区域作为目标模板进行模板匹配跟踪,提高了目标模板提取精度。实验结果表明,相对于传统采用矩形模板的匹配跟踪方法,该方法具有更高的抗背景干扰能力,有效提高了对复杂背景下运动小目标的跟踪稳定性。(2) 围绕目标跟踪中常遇到的因光照变化、遮挡以及表观变形导致的跟踪稳定性下降问题,设计了一种模板自适应更新策略,在此基础上提出了一种基于轮廓模板匹配的目标跟踪方法。设计并实现了具有自适应模板更新策略的孪生跟踪网络,系统从单点初始化生成目标轮廓模板,跟踪过程中由目标检测分支与轮廓检测分支共同决策适时更新模板,以达到最佳性能。通过实验证明该方法可有效解决目标部分遮挡以及形变问题,中心误差统计结果表明,该方法的跟踪精度较其他方法有较大幅的提升。(3) 提出了一种基于多传感器信息融合的捷联成像导引视线角速度信号求解方法。建立了捷联成像导引视线角速度计算模型,根据视觉通道的跟踪结果和陀螺仪得到的姿态信息求解目标视线角速度信号。针对移动智能终端的陀螺存在零点漂移、随机噪声以及精度低,目标跟踪算法存在跟踪误差等问题,通过无迹卡尔曼滤波来提高视线角速度的解算精度。仿真实验结果表明,本文提出的方法可以一定程度上克服陀螺等惯性传感器零漂大、精度低的问题。(4) 采用华为移动智能终端HUAWEI Mate 9完成了捷联成像导引头的开发,基于Mate 9的处理器、视觉传感器、惯性传感器、卫星导航模块以及操作系统,完成了目标跟踪、视线角速度解算以及视线角速度滤波算法的开发,并进行了试验验证,为低成本、微型捷联成像制导组件的研制提供了一种可行的技术途径。
Other AbstractThis paper takes the tracking and guidance of small moving targets in the complex background as the application scenario,and carries out the research of Strap-down imaging terminal guidance technology based on the mobile intelligent device. Using the low-cost sensor and processor components of the mobile intelligent terminal, the design, development and technical verification of Strap-down imaging guidance components are completed. The main research work is summarized as follows: (1) This paper studies the relationship between the selection of target template and the tracking stability around the problem of small moving target tracking in complex background. On this basis, in order to improve the stability of small moving target tracking in complex background, a target template extraction method based on deep learning is proposed. Firstly, a deep contour detection network is designed to extract the contour of the target. A number of experiments show that the depth contour detection network designed in this paper can extract the contour information better than other methods. Then, the area enclosed by the target contour is selected as the template for template matching and tracking, which improves the accuracy of template extraction. The experimental results show that compared with the traditional matching tracking method with rectangular template, this method has higher anti background interference ability and effectively improves the tracking stability of small moving targets in complex background. (2) In this paper, a template adaptive updating strategy is designed to solve the problem of tracking stability degradation caused by illumination change, occlusion and affine deformation, which is often encountered in target tracking. On this basis, a target tracking method based on contour template matching is proposed. The Siamese tracking network with adaptive template updating strategy is designed and implemented. The system generates the target contour template from a single initialization point. In the tracking process, the target detection branch and the contour detection branch jointly decide to update the template in time to achieve the best performance. The experimental results show that this method can effectively solve the problem of partial occlusion and deformation of the target. The statistical results of the central error show that the tracking accuracy of this method is greatly improved compared with other methods. (3) This paper presents a method to solve the LOS angle velocity signal of Strap-down imaging guidance based on multi-sensor information fusion. The calculation model of LOS angle velocity of Strap-down imaging guidance is established, and the LOS angle velocity signal of target is calculated according to the tracking result of vision channel and the attitude information obtained by gyroscope. In view of the characteristics of large gyro zero drift of mobile intelligent devices and the influence of background noise and thermal noise, the accuracy of LOS angle velocity is improved by unscented Kalman filter. The simulation results show that the method proposed in this paper can overcome the problem of large zero drift and low accuracy of pose sensor to some extent. (4) A Strap-down imaging seeker is developed by Huawei mobile intelligent device Huawei mate 9, which is based on mate 9's processor, vision sensor, inertial sensor, satellite navigation module and operating system have completed the development of target tracking, LOS angle velocity calculation and LOS angle velocity filtering algorithm, and have carried out experimental verification, which provides a feasible technical way for the development of low-cost, micro Strap-down imaging guidance module.
Language中文
Contribution Rank1
Document Type学位论文
Identifierhttp://ir.sia.cn/handle/173321/27155
Collection光电信息技术研究室
Affiliation中国科学院沈阳自动化研究所
Recommended Citation
GB/T 7714
徐峥. 基于移动终端的捷联成像末制导技术研究[D]. 沈阳. 中国科学院沈阳自动化研究所,2020.
Files in This Item:
File Name/Size DocType Version Access License
基于移动终端的捷联成像末制导技术研究.p(4703KB)学位论文 开放获取CC BY-NC-SAApplication Full Text
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[徐峥]'s Articles
Baidu academic
Similar articles in Baidu academic
[徐峥]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[徐峥]'s Articles
Terms of Use
No data!
Social Bookmark/Share
All comments (0)
No comment.
 

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.