水面无人艇的多传感器信息融合环境理解与建模技术研究

The accurate local environment comprehending and modeling of an unmanned surface vehicle (USV) is a fundamental supporting technology of its autonomy, which is also the crucial technical guarantee for an USV to accurately complete intelligent behaviors such as path planning, situation assessment, and threat avoidance. However, at present the environment comprehending and modeling technology of USVs still could not completely support mission requirements, and the foreign and domestic related research work is still in its primary stage. In the proposed project a multi-sensor environment sensing system of USVs is constructed by a marine radar, a camera and an infrared imager, and research work on radar multiple moving target tracking and environment modeling, target and environment comprehending based on optical sensors, multi-sensor information association and registration of environment and targets, and multi-sensor environment information fusing and modeling is carried out to implement accurate local environment comprehending and modeling of USVs. In the end the environment comprehending and modeling method will be tested in simulation and typical marine environment, and the accuracy, reliability and real-time performance will be verified and assessed. The research work in the proposed project could improve navigation safety and mission execution autonomy of USVs in undefined non-structured marine environment.

水面无人艇（USV）对局部环境的准确理解与建模是其自主性技术的基础支撑技术，也是USV准确完成诸如路径规划、态势评估、危险规避之类的自主行为的关键技术保障。目前USV的环境理解与建模技术还未能够完全支持任务需求，国内外相关研究工作主要处于初级阶段。本项目以雷达、摄像机、红外成像仪构成USV的多传感器环境感知系统，拟开展雷达多运动目标跟踪与环境建模技术、基于光学传感器的目标与环境信息理解技术、环境与目标的多传感器信息关联与匹配技术、多传感器环境信息融合与模型构建技术的研究工作，实现对与USV本体关联的局部环境的准确理解和建模，在仿真环境和典型海洋环境中完成验证试验，对系统的准确性、实时性、可靠性进行验证和评估，提高USV在非结构化的不确定海洋环境中航行的安全性和执行使命任务的自主性。

以水面无人艇（Unmanned Surface Vehicle, USV）利用航海雷达、高清摄像机、红外成像仪等传感器实现对非结构化不确定海洋环境的理解和建模为研究目标，建立了多传感器信息融合环境理解与建模技术的硬件和软件结构，开展了雷达多运动目标跟踪与环境建模技术、基于光学传感器的目标与环境信息理解技术、环境与目标的多传感器信息关联与匹配技术、以及多传感器环境信息融合与模型构建技术的研究工作，在仿真环境和典型海洋环境中完成了验证试验，实现了对USV自主环境理解和建模能力的客观评价，为USV在复杂海洋环境下自主完成各种使命任务奠定了重要技术基础。. 主要针对导航雷达存在海杂波干扰、扫描周期长、目标回波位置跃变等因素带来的目标检测性能下降问题，在大数据理论基础上研究并提出海杂波分布规律预测模型，实现对运动目标的检测和静态环境的建模，最后采用容积卡尔曼滤波方法根据运动目标的历史状态和当前量测信息实现对多运动目标的持续跟踪。. 针对光学图像中海面散射光和天空背景辐射干扰因素，研究海面光学成像传输规律并用于图像信息重建，同时建立背景干扰预测模型，对光学图像中的背景干扰进行抑制，研究海天线/海岸线快速检测方法，然后实现运动目标的快速检测和持续稳定跟踪，完成海面目标的定位和静态环境理解。. 有效的多传感器信息融合要求能够进行传感器联合立体标定，提取传感器信息中环境和目标的关键点等显著特征，利用关键点在各传感器中的映射关系建立传感器之间的几何空间约束条件，利用该约束条件可完成传感器信息的快速匹配，实现对目标更精确的定位和环境信息更准确的理解。. 研究异构多传感器的特征深度融合技术，利用语义特征完成特征的匹配、融合，进行不依赖于传感器物理特性的高层语义特征融合，同时包含基本的空间结构信息，具有较高的鲁棒性、可靠性，进行目标状态的最优估计，结合艇体运动状态实现环境模型的自主构建和实时更新。