• 中国计算机学会会刊
  • 中国科技核心期刊
  • 中文核心期刊

计算机工程与科学

• 论文 • 上一篇    下一篇

基于前后双摄像头的鲁棒视觉里程计算法

施啸天1,张宇2,方舟1   

  1. (1.浙江大学航空航天学院,浙江 杭州 310027;2.浙江大学控制科学与工程学院,浙江 杭州 310027)
  • 收稿日期:2016-04-25 修回日期:2016-07-01 出版日期:2017-10-25 发布日期:2017-10-25
  • 基金资助:

     基金项目:浙江省自然科学基金(LY15F030005);浙江省公益技术应用研究计划(2016C33246);国家自然科学基金

     

     

    (61673341);中央高校基本科研业务费专项资金(2016QNA5010)

Robust visual odometry based on front and back cameras
 

SHI Xiao-tian1,ZHANG Yu2,FANG Zhou1   

  1. (1.College  of Aeronautics and Astronautics,Zhejiang University,Hangzhou 310027;
    2.College of Control Science and Engineering,Zhejiang University,Hangzhou 310027,China)
     
  • Received:2016-04-25 Revised:2016-07-01 Online:2017-10-25 Published:2017-10-25

摘要:

针对室内视觉里程计受光照影响明显和现代移动设备双摄像头的特点,提出同时利用双侧摄像头视觉里程计算法。通过对单
侧视觉里程计状态的评估和重启,实现单侧故障时系统的持续稳定输出,提高了视觉里程计的鲁棒性。在双侧正常运行时通
过对双侧视觉里程计输出结果的卡尔曼滤波融合,提高了视觉里程计的精度。在室内多种实际环境、多种移动速度下的实验
结果表明,本算法有效地保证了单侧异常情况下视觉里程计的正常输出,同时在双侧有效的情况下明显提高了里程计输出精
度。

关键词:

Abstract:

We propose a new visual odometry method to use front and back cameras at the same time due to the terrible
impact of luminance on indoor visual odometry and the dual-camera feature of modern  mobile devices. By the
state estimation of the visual odometry on each side, we can achieve the sustained output even when one side
fails, and the robustness of visual odometry is improved. In addition, we fusion the output of the two-side
visual odometry by Kalman filter to improve the accuracy when the two sides both run normally. Experimental
results under various practical environments in various velocities show that the system can output regularly
even when one side fails and the accuracy of the output is improved when the two sides work normally.

Key words: visual odometry, Kalman filter, information fusion, smart phone