Registration of Depth Image and Color Image Based on Harris-SIFT

In the recent past, it has become a widely accepted and applied method to recognize and localize objects based on local point features. However, the existing approaches are based on matching color images. With the widely use of Kinect, the registration between depth image and color image becomes necessary and important. To effectively match 3D depth images and 2D color images and estimate the transformation homography, we present a type of features, which combines Harris corner detector with SIFT descriptor. In the experiments, the corresponding feature points between depth image and color image can be well located by the descriptor, which is invariant under blurring, rotation, shift, scaling and moderate changes in viewpoint. Keywords-Harris-SIFT; depth image; registration