三维模型边缘特征与异常检测

Edge features and anomaly detection for 3D models

针对三维模型识别和检测问题,提出一种新的基于边缘特征的三维模型异常检测方法。将每一个三维模型利用边缘特征表示为一条时间序列,对产生的时间序列集进行Isodata聚类,利用聚类结果经过两次划分实现异常检测。第一次划分过程产生候选异常和候选正常,第二次划分过程在候选异常中进一步选出检测结果。实验结果表明,该算法性能优于传统的基于距离、邻近度以及基于相对密度的异常检测算法,在一定条件下,也优于基于密度的异常检测算法。

Aiming at problems in identification and detection of 3D models,a new method of anomaly detection based on edge features for 3D models is presented.Each throe-dimensional model is expressed as a time series through the edge features.Then the obtained time series dataset are clustered using the isodata algorithm.Anomalies are detected by partitioning the dataset twice using the clustering results.It partitions the dataset into two subsets,the preparatory norm set and the preparatory anomaly set, and then the final anomalies are further filtered from the preparatory anomaly set.Experiment results show better performance of the proposed method compared with anomaly detection methods based on distance, neighbourhood or relative density.Under certain conditions,it is also better than density based anomaly detection.