一种基于支持向量机的孔洞修补方案评估方法

A Hole Recognition Method Based on the Optimized SVM

为了自动评估逆向设计软件的点云孔洞修补效果,以确定最佳的修补方案,提出了一种基于支持向量机的点云孔洞修补评估算法。首先,以基于实际工程应用的挖掘机斗齿点云为实验对象,在斗齿的不同部位人为构造20个点云孔洞,用专业逆向设计软件Imageware13.2进行孔洞修补,并建立相关特征向量。其次,将60%的特征向量作为训练样本输入SVM模型进行训练,并建立参数优化后的SVM分类器模型。最后,将其余特征向量作为测试样本输入SVM模型进行分类识别。实验证明,此方法具有较好的识别效果,识别正确率可达90%,并有鲁棒性好、识别速度快等优点。

This paper presents a novel hole-recognition algorithm in reverse engineering （ R E ） that can automatically evaluate the effectiveness of the software based hole filling method thus to determine the best filling schemes for holes in the point clouds. First, 20 manmade holes were circle selected from different portion of a real-world scanned objects-a bucket of an excavator, and then all of them were reconstructed with Imageware 13.2. Second, 6 0 % of them were used to train the S V M and to establish the pattern recognition model, while others were for S V M predicting. The experimental results show that the suggested approach performs quite well, it is efficient, robust and what＊ s more, the average recognition accuracy can be up to 87.5%.