基于双目视觉的无人船障碍物探测与地图构建

Obstacle detection and map construction of unmanned boat based on stereo vision

摘要:

在无人船的巡航过程中，如何探测和躲避障碍物是必须解决的问题。由于无人船自主巡航依赖于其自身对所处环境的精确探测，而目前一般采用的超声波或雷达技术探测距离和精度较低，避障功能弱，因此需引入视觉方法来提升避障精度，并用于轨迹生成、定位或路径规划。本文为无人船搭建双目视觉系统，提出采用双目直接稀疏里程计算法构建无人船所处三维空间模型，将构建的三维点云图转化为二维网格图并标记障碍物，为避障系统进行路径规划提供障碍物数据。实验结果表明，该系统对真实河流环境进行了地图构建，并解决了水上环境构建存在的“虚拟障碍”等问题。

Abstract:

Obstacle detection and recognition during navigation is a key challenge for unmanned boat. Since the autonomous cruise of unmanned boat depends on their own accurate detection of their environment, and the ultrasonic or radar technologies commonly used at present have low detection distance and accuracy and weak obstacle avoidance function, so visual methods can be introduced to improve the obstacle avoidance accuracy and used for trajectory generation, localization or path planning. In this paper, stereo vision system is built for unmanned boat, and the 3D space model of the boat is constructed by using stereo Direct Sparse Odometry(DSO) algorithm. Then the 3D point cloud map will be transformed into a 2D grid map and the obstacles will be marked. At last, the obstacle data for obstacle avoidance system is provided to plan path. Experimental results indicate that the system constructs a 2D grid map of the real river environment and solves the problem called "virtual obstacle" in the construction of water environment.