基于单目视觉的水面目标识别与测距方法研究

无人船在自主航行过程中,如何探测障碍物种类与距离是值得重点研究的问题。现阶段的无人船主要采用激光雷达、超声波等设备探测水面障碍物,但此类方法并不能识别出障碍物的具体类型。提出采用基于单目视觉的目标检测与测距算法,不仅能够检测出无人船前方是否有障碍物,还能够对水面障碍物进行识别和距离检测。首先通过训练好的Mask R-CNN模型对水面障碍物进行检测识别,随后利用改进的基于小孔成像原理的单目测距方法对识别到的障碍物测量距离。为验证该目标识别与测距方法的有效性,选取了不同形状的船只、河岸、荷叶及荷花等图像进行验证实验。实验结果表明,课题研究的目标识别与测距方法能较为准确地对水面障碍物进行分类及像素级别的实例分割,同时测量障碍物与无人船之间的距离。

Research on Water Surface Target Recognition and Ranging Based on Monocular Vision

In the process of autonomous navigation,how to detect the type and distance of obstacles is worth studying.At present,unmanned ships mainly use lidar,ultrasonic and other equipment to detect obstacles on water surface,but such methods cannot identify the specific types of obstacles.The target detection and ranging algorithm based on monocular vision is proposed,which can not only detect whether there are obstacles in front of the unmanned ship,but also identify the obstacles on the water surface and detect the distance.Firstly,the trained Mask R-CNN model is used to detect and identify the obstacles on the water surface,and then the improved monocular ranging method based on keyhole imaging principle is used to measure the distance of the identified obstacles.In order to verify the effectiveness of the target recognition and ranging method,the images of water surface obstacles such as ships,riverbanks and lotus with different shapes are selected for verification experiments.The experiment shows that the target recognition and ranging method studied can classify the water surface obstacles and segment the examples at the pixel level more accurately,and measure the distance between the obstacles and the unmanned ship at the same time.