基于自适应聚类与YOLOV4的内河靠泊点云识别方法

Identification Method of Inland River Berthing Point Cloud Based on Adaptive Clustering and YOLOV4

针对智能船舶内河靠泊过程中面临的常见障碍物目标检测与精准测距需求,提出一种基于点云自适应聚类与YOLOV4结合的目标点云与空间距离测量方法。首先,采用体素滤波对原始3D点云数据进行预处理,减少计算量;其次,采用K值自适应聚类方法提取目标特征;最后,将处理后点云转化为具备深度信息的鸟瞰图,利用YOLOV4算法识别出内河靠泊过程中的目标物类型,并进行定位。实船试验结果表明:该方法可作为实用辅助系统,实时识别出趸船、靠泊船舶等目标,并输出距离信息;该方法对趸船、泊位等目标识别精度达到了96.93%,测距精度达到厘米级别,将为未来自主靠泊提供重要支撑。

Aiming at the common obstacle target detection and accurate ranging requirements faced by intelligent ships in the process of berthing and unberthing in inland rivers,a target point cloud and spatial distance measurement method based on the combination of point cloud adaptive clustering and YOLOV4 was proposed.Firstly,voxel filtering was used to preprocess the original 3D point cloud data to reduce the amount of calculation;secondly,the K-value adaptive clustering method was used,which has better performance in complex scenes such as berthing and berthing,and makes the target features more obvious;finally,the point cloud was converted into a bird′s-eye view with depth information,and the YOLOV4 algorithm was used to identify the type of objects in the process of berthing and unberthing in inland rivers,and located them.The real ship test results showed that this method can be used as a practical auxiliary system to identify targets such as barges and berthing ships in real time and output the distance information.The method can achieve 96.93%recognition accuracy for pontoons,berths and other targets,and the ranging accuracy has reached the centimeter level,which will provide important support for autonomous berthing and berthing in the future.