Robust 3D keypoint detection method based on double Gaussian weighted dissimilarity measure

Abstract

This paper proposes a novel multiscale 3D keypoint detection method via the double Gaussian weighted dissimilarity measure. At each scale, the shape index value and the double Gaussian weighted dissimilarity measure value of each 3D point are firstly computed. Then the candidate keypoints with local maximum dissimilarity measure values are determined. Finally the final 3D keypoints are detected under our proposed multiscale detection scheme. As the dissimilarity measure used in this paper has better robust descriptive ability and is rotation and translation transformation invariant, the proposed detection method is robust to noise, rotation and translation transformation. Extensive experimental results have shown that using our proposed multiscale detection method, we can detect the keypoints with higher repeatability under different noise levels.