基于几何特征的曲面物体识别方法是通过从景物深度图象上提取景物表面的高斯曲率和平均曲率、曲率直方图,曲率的熵等几何信息.将景物用一个属性关系图ARG来表示,并与模型库中的模型ARG图进行优化匹配,从而来识别曲面景物.该方法主要是针对机器零部件等人造曲面物体的识别问题而设计的,其曲面几何特征的描述方法对二阶曲面比较有效,实验表明,应用该方法可成功地从深度图象中识别机器零部件等曲面物体,且有较好的识别结果.文中所描述的曲面物体表示方法,可进而推广到较复杂的曲面描述及较复杂的三维曲面物体的识别.

In this paper, a method for the recognition of curved object is presented which is based on the geometrical features of object. First, geometrical information including Gaussian culvature, mean culvature, histogram of culvature, entropy of culvature, etc. is extracted from scene range image, then the scence is represented as a attribute relational graph(ARG) and is matched optimizely with model ARGs of model database. The approach is designed aimming at the recognition of artificial curved objects such as machine parts and components, etc, the curved surface represent of that curved geometrical features is efficient for second order curved surface. The method has been simulated on a PC computer (PentiumII), and have got some good results. The results indicated that curved objects(such as machine parts and comonents) could be successfully recognized from range images with this method.The represent of curved surface in this paper could be extented to the represent of more complex curved surface, and the approach could also be extended to the recognition of more complex 3D curved object.