单目视觉引导下的无人艇局部避障方法

针对无人艇（unmanned surface vehicle，USV）在作业过程中，需要及时躲避未被标注在地图上的漂浮障碍，并及时返回规划航道的问题，提出一种基于单目视觉的无人艇局部避障方法。避障方法使用YOLO网络识别图像中的障碍物，结合针孔成像原理和水体自由断面特点，解决单目视觉尺度不确定问题，从图像中恢复出障碍物的宽度及深度信息；针对无人艇惯性大，机动性差的特点，引入虚拟水道宽度等信息素，提出了改进的向量场直方图算法（vector field histogram plus，VFH+）；算法参考当前障碍疏密，动态调整不同信息素对最佳航向角计算的影响，引导无人艇在兼顾全局导航信息情况下，应对未知漂浮障碍物快速自动局部避障。基于水动力模型和Gazebo仿真引擎搭建无人艇避障的仿真实验平台。在仿真中快速迭代研究算法的参数范围及控制效果，实现了无人艇在舱内空间和能量均有限的工况下，仅使用单目视觉传感器，兼顾全局路径规划和局部障碍信息，自动化高效躲避漂浮障碍，并在脱离障碍后回归原有规划路径的功能。

Local obstacle avoidance for unmanned surface vehicle via monocular vision

In order to solve the problem that unmanned surface vehicle (USV) needs to avoid unmarked floating obstacles in the process of movement, a local obstacle avoidance algorithm is designed based on monocular vision and keeps the USV in the desired way. In the algorithm, YOLO is used to recognize the location of the obstacles in the image. Orientation and depth information of the obstacles can just be derived from the monocular vision without considering its uncertainty scale problem via the principle of pinhole camera and the characteristics of the free surface of the water. Due to the large inertia and low flexibility of the USV, an improved vector field histogram plus (VFH+) algorithm, introducing the factor of the width of the virtual waterway, dynamically adjusting the influence of different factors via the density of the obstacles to search the optimal heading angle, can automatically guide the USV to avoid unknown floating obstacles considering global path planning information. Based on the hydrodynamic model and the Gazebo simulation engine, we build a simulation experimental platform for obstacle avoidance of unmanned vehicle. After quickly and iteratively searching the parameter range and comparing the effect of the algorithm in simulation, the USV with limited space and energy can efficiently and automatically avoid floating obstacles by using only monocular vision sensors, considering both the global path planning and local obstacle information, and return to the original planned path after breaking away from obstacles.