Fig. 2From: Novel 3D local feature descriptor of point clouds based on spatial voxel homogenization for feature matchingIllustration of descriptor principle. (a) Original 3D object, where yellow points represent the keypoints obtained from the surface of the object; (b) The local surface within neighboring space with radius r and centered on keypoint p; (c) Construction of LRF on keypoint p and transformed local points with respect to the LRF, where blue, yellow, and red arrows represent X-, Y-, and Z-axis, respectively; (d) Split space into a set of cubic voxels; (e) Set label for each voxel, where blue voxels and transparent voxels represent label value is 1 and others is 0, respectively; (f) Update the label value of each voxel, where the color depth of the cube represents the size of voxel label value; (g) Output feature of keypoint p by merging label value of each voxel according to particular index codingBack to article page