大场景下港口装卸设备点云的聚类识别算法研究

Research on Clustering Identification Algorithm for Point Cloud of Port Handling Equipment

针对港口装卸设备位姿变化明显的特点,提出了一种基于区域生长思想的K-Normal聚类算法。该算法在区域生长聚类思想基础上,在进行K邻域搜索时分段进行随点云局部密度改变的距离阈值判定以及法向量夹角判定完成准确聚类,能有效避免过分割与欠分割。聚类完成后通过PCA算法计算不同聚类的特征值,构建特征模型作为全局特征,以特征模型为输入构造SVM分类器,完成港口不同目标的识别,识别正确率达90.5%。

The position and pose of port handling equipment changes obviously.Based on region growing,a clustering algorithm named as K-Normal to process point cloud data of port handling equipment is proposed in this paper.The algorithm combined the distance threshold judgment with the angle threshold judgment of normal vector during KNN searching to avoid over-segmentation and under-segmentation.After clustering,PCA was used to calculate the characteristic values of different clusters,and the feature model was constructed as the global feature.A SVM classifier was constructed which the feature model acts as the input to complete the identification of different targets in port.The accuracy of algorithm is up to 90.5%.