基于深度学习的三维感知算法在船舶自主靠泊场景中的应用

Application of 3D perception algorithm based on deep learning in ship autonomous berthing scene

随着船舶自主化在航运业的快速发展,《自主货物运输船舶指南》提出了针对自主靠泊场景的环境感知等相关技术要求。本文提出一种基于深度学习的三维感知算法,将多传感器的点云和图像数据进行深度融合,提升目标检测和分类的精度,用于寻找指定的靠泊目标。搭建基于虚拟物理引擎的船舶靠泊仿真系统,模拟靠泊场景,解决数据采集的难题。最后结合成熟的决策和控制算法,构建完整的自主靠泊系统,完成靠泊仿真试验,验证算法的有效性,对船舶自主靠泊应用具有重要意义。

With the rapid development of ship autonomy in the shipping industry,the “Guidelines for Autonomous Cargo Ships”proposed related technical requirements such as environment perception for autonomous berthing scenarios.This paper proposed a 3D perception algorithm based on deep learning,which used deep fusion of multi-sensor point clouds and image data to improve the accuracy of target detection and classification,and was used to find specified berthing targets.A ship berthing simulation system based on a virtual physical engine was built to simulate berthing scene and solve the problem of data collection.Finally,combined with mature guidance and control algorithms,a complete autonomous berthing system was constructed,and completed the berthing simulation experiment,which verified the effectiveness of the proposed algorithm.It had significant value for autonomous berthing application.