Wide-Angle Visual Feature Matching for Outdoor Robot Navigation

We describe an outdoor mobile robot navigation and localisation system based on a new visual feature detector designed for wide-angle image sequences. Wide-angle images exhibit significant radial distortion for which existing scale-space detectors such as SIFT are inappropriate. We argue that the required scale-space images can be correctly obtained through the convolution of the image, mapped to the sphere, with the spherical Gaussian. Several computational approaches to this convolution are investigated in the spatial and frequency domain. In particular a close approximation is developed that has comparable computation time to conventional SIFT but with improved matching performance. We present results from two monocular wide-angle outdoor image sequences obtained using fisheye and equiangular catadioptric image sequences. We evaluate the overall matching performance (recall vs 1-precision) of these methods compared to conventional SIFT. We also demonstrate the use of the technique for variable baseline visual odometry and for place recognition using algorithms inspired by “Video Google”.