Improving viewpoint invariance of image feature extraction methods using intensity and range images

The most common image feature extraction algorithms such as SIFT (Scale Invariant Feature Transform) and SURF (Speeded Up Robust Feature) have been proven to be invariant to changes in rotation, scale and with restrictions to illumination and viewpoint changes. These algorithms generate descriptor vectors around keypoints in 2D images. Close descriptors suggest similar image patch. In case of mobile robotics applications it is important to achieve good viewpoint invariance and stability to detect landmarks and objects with high reliability. Improving viewpoint invariance for image feature detection increases the efficiency of SLAM algorithms. In this paper we present and evaluate a method to use additional data provided by range image sensors to supplement traditional feature extraction algorithms to improve viewpoint invariance. We present the method and results of computer simulation and also real world examples comparing the SURF (OpenSURF) with and without the improvement. An active structured light based range and intensity image sensor was used to acquire real world test images.