This paper suggests an indoor environment descriptor and global localization strategies for indoor robot navigation using a metric sensor and mono camera. Other researches use various feature descriptors (i.e. geometric features, visual local invariant features, and objects) for robot pose estimation. However, most of the real environments have repeated similar texture patterns or few objects although they need salient information for successful localization. To overcome this problem, we suggest a new environment descriptor, which is composed of the histogram of oriented gradient(HOG) and approximated 2D-polar coordinate distance of visual vertical edges. We call it Distance-assisted HOG (DaHOG). For the matching process, we use the omnidirectional datasets that have a circular pattern matching problem. Here, we solve the problem by proposing a new global localization method based on a spectral matching technique. We show that our method is effective with experiments in real environments where there is a lack of distinctive features and objects.