Local Ordinal Intensity And Normal RGB-D Descriptor

In the fields of computer vision and robotics, feature matching is the essential operation in many complex problems, such as image stitching, wide baseline matching, coarse alignment and image retrieval etc. With the population of RGB-D sensors, there are some new works focusing on combining both RGB and Depth information to construct uniform descriptors. However, current RGB-D combination based descriptors also have some limitations, such as dependence on the accuracy of the geometric information and high computational complexity.In this paper, we introduce a novel RGB-D descriptor called local ordinal intensity and normal descriptor (LOIND) with the integration of texture information in RGB image and geometric information in depth image. The main contributions are as follows:1. We implement a descriptor with a 3-D histogram supported by orders of intensities and angles between normal vectors, in addition with the spatial sub-divisions. The former ordering information which is invariant under the transformation of illumination provides the robustness of our descriptor, while the latter spatial distribution provides higher information capacity.2. A timesaving method for scale transformation is proposed. Unlike other scale estimation method like Gaussian Pyramid, we can estimate the scale of each feature point based on the geometrical information absolutely.3. We also provide a new method to estimate the dominant orientation with only the geometric information, which can ensure the rotation invariance under extremely poor illumination condition.