论文部分内容阅读
利用单幅图象中物体的三条边与模型中的三条对应边,可求出三维物体姿态,但解不唯一,通过将这些可能姿态所产生的图象与实际图象匹配,可求出唯一正确姿态.二维图象特征对应问题是个NP完全问题,存在组合爆炸的困难,为此,我们把特征对应问题看作一个组合优化问题,利用Hopfield网络成功解决这一组合优化问题.该算法通用性强,而且适合于并行实现。文中给出了在Ⅵ-COM图象处理系统上对人造图象和实际图象进行的实验结果。
Using the three sides of an object in a single image and the three corresponding edges in the model, the three-dimensional object’s pose can be found, but the solution is not unique. By matching the images generated by these possible poses with the actual image, the unique Correct pose.For two-dimensional image feature correspondence problem is a NP complete problem, there is a difficulty of combinatorial explosion, so we regard the feature correspondence problem as a combinatorial optimization problem, using Hopfield network successfully solve this combinatorial optimization problem. Strong, but also suitable for parallel implementation. The experimental results of artificial images and real images on VI-COM image processing system are given in this paper.