A registration method based on angular-invariant feature

This paper proposes an angular invariant feature for registration procedure to perform reliable matching between a given model shape and a data shape. The feature is defined by a k-dimensional vector with k angles between the normal vector of each point and its k-nearest-neighbors individually. The feature is invariant to scale transformation, rotation transformation. Particularly, when we register two surfaces with small curvatures, in that, nearly-flat meshes, it is difficult to reach the correct alignment because of fewer features in traditional ICP algorithm and its variants. However, the method presented here can discover the potential structural information hiding in nearly-flat surfaces and improve the convergence and error. We present that an angular augment coefficient should be set to avoid rounding error of computer. This new invariant feature makes defect detection of surfaces, such as airplane, faster and more robust.