Application Research On Point Cloud Clustering And Identification Of General Cargo Wharf Handling Equipment

Driven by the policy of "Guidance on the construction of world-class ports",the construction of smart ports is rapidly advancing.As an important part of the port,the general cargo terminal is in urgent need of automation and unmanned construction,and it is a key step to automatically obtain the position and posture of various loading and unloading equipment.At present,the traditional pose acquisition scheme of the loading and unloading equipment of the general cargo terminal is prone to cumulative error,and the image vision is easily disturbed by environmental factors.Position and pose acquisition based on LIDAR point cloud data can simultaneously obtain multiple types of position and pose information and can be used all day long,but its application in ports is still in its infancy.Based on point cloud data processing,this paper researched on identifying point cloud of different handling equipment of general cargo.The main research contents of this paper are as follows:(1)In view of the characteristics of many and miscellaneous discrete points of point cloud in port open environment,the algorithms of down-sampling,outlier elimination and ground point segmentation are compared and selected,parameters are determined and the algorithm is improved,so as to improve the filtering rate of point cloud quantity and speed up point cloud pretreatment on the premise of retaining the characteristics of point cloud.(2)In view of the characteristics of large variation range of position and pose of general cargo terminal loading and unloading equipment,the disadvantages of the existing DBSCAN algorithm in dealing with point clouds with large variation of density,and the disadvantages of over-segmentation and under-segmentation in dealing with dense scenes are analyzed.Based on the idea of regional growth and clustering,this paper proposes a K-Normal clustering algorithm based on the threshold of changing distance and the angle between normal vectors.The algorithm segmented the K neighborhood,combined with the threshold of distance changing with the local density of point cloud and the Angle between Normal vectors to achieve accurate clustering.(3)The recognition algorithms of local features and global features are compared and analyzed.PCA algorithm is used to construct global features and SVM classifier is combined to complete the recognition of point clouds of different loading and unloading equipment.When constructing global features,the function std is constructed to select the optimal model,and SVM classifier is constructed with the optimal feature model as input.lib SVM and voting mechanism are used to identify different targets of the port.In this paper,LIDAR is used for the identification and pose acquisition of the loading and unloading equipment of the general cargo terminal.The applicability of various point cloud processing algorithms to this paper is analyzed and compared,and improved and innovated to design a system suitable for the identification and pose acquisition of the loading and unloading equipment of the general cargo terminal.The system can effectively avoid over-segmentation and under-segmentation in the process of point cloud clustering,and the recognition accuracy is up to 90.5%.The algorithm processing speed meets the real-time requirements,which is of substantial significance to promote the automation and unmanned development of the cargo terminal.