Underwater navigation is always a challenging problem because of electromagnetic attenuation. The traditional methods involve beacons, inertial sensors, and Doppler velocity log, but they have many shortcomings, such as high cost and lengthy setup time. In order to solve underwater navigation problems at low cost, an integrated visual odometry system has been developed and discussed in this article. In this method, two inertial sensors provide acceleration and attitude of the vehicle, and an underwater sonar is used to provide the distance between the vehicle and the seabed, whilst in the visual odometry section, an optical flow algorithm has been applied for tracking feature points. With the depth provided by the sonar, 3-D position of feature points can be calculated. Linear motion of the vehicle is then predicted through these feature points in dual frames. Finally, nonlinear optimization is used to correct the attitude of the vehicle using visual information. In the proposed algorithm, the vehicle trajectory can be estimated in absolute scale by using a single camera; computational complexity is reduced dramatically compared to other visual odometry methodologies; and this algorithm allows the approach to work in sparse texture conditions. The results from practical experiments demonstrate that the method is effective and it is also a low-cost solution.

中文翻译：

---

水下航行器综合视觉里程计系统

由于电磁衰减，水下导航始终是一个具有挑战性的问题。传统方法涉及信标、惯性传感器和多普勒速度测井，但存在成本高、设置时间长等缺点。为了以低成本解决水下导航问题，本文开发并讨论了一种集成视觉里程计系统。在该方法中，两个惯性传感器提供车辆的加速度和姿态，并使用水下声纳提供车辆与海床之间的距离，而在视觉里程计部分，已应用光流算法来跟踪特征点. 利用声纳提供的深度，可以计算特征点的 3-D 位置。然后通过双帧中的这些特征点预测车辆的线性运动。最后，非线性优化用于使用视觉信息校正车辆的姿态。在所提出的算法中，可以使用单个摄像头在绝对尺度上估计车辆轨迹；与其他视觉里程计方法相比，计算复杂度显着降低；并且该算法允许该方法在稀疏纹理条件下工作。实际实验结果表明，该方法是有效的，也是一种低成本的解决方案。与其他视觉里程计方法相比，计算复杂度显着降低；并且该算法允许该方法在稀疏纹理条件下工作。实际实验结果表明，该方法是有效的，也是一种低成本的解决方案。与其他视觉里程计方法相比，计算复杂度显着降低；并且该算法允许该方法在稀疏纹理条件下工作。实际实验结果表明，该方法是有效的，也是一种低成本的解决方案。