从单幅图像定性识别多面体的代数方法

AN ALGEBRAIC APPROACH FOR 3-D OBJECT RECOGNITION FROM A SINGLE VIEW

本文提出了一种三维目标定性识别的方法,利用可见表面的NRLCC码建立目标与模型的初始匹配,然后利用一种代数方法进行验证,其中引入了一个广义尺度因子λ_i,i=1,…,n,使得识别对于部分丢失边界以及透视投影引起的中等程度的失真均不敏感.这种算法本身简洁明了,避免了优化过程的复杂运算,同时又能获得鲁棒的识别效果.

An approach for qualitative 3-D object recognition is presented. First, an initial matching between the object and the model is established using NRLCC code of a visible surface. Then generalized scalar factors i, i = 1, ...., n. are introduced in the verification process, which enables the approach insentsitive to partially missing edges and moderate distortion introduced by perspective projection. This approach is very simple and avoids the time-consuming numerical procedure. At the same time, we can obtain robust recognition result.