Ego-localization using streetscape image sequences from in-vehicle cameras

This paper focuses on ego-localization using in-vehicle cameras. We propose a 2D ego-localization method using streetscape appearance as a feature of image matching and the triangulation of matching results. The image sequences of two in-vehicle cameras are matched to a database that contains a sequence of streetscape images and their corresponding positions. First, the proposed method searches for images similar to the input image from the database. Second, vehicle position is calculated based on triangulation using the positions stored in the database and the viewing directions of the two cameras. By assuming that the streetscape appearance changes continuously, a sequential image matching algorithm is used to improve the ego-localization accuracy. From experimental results, we confirmed that the proposed method surpasses the accuracy of a general GPS and achieved sufficient accuracy to be used for driving lane recognition.