Drift Compensation of Mono-Visual Odometry and Vehicle Localization Using Public Road Sign Database

摘要

This paper proposes a novel localization method based on a camera that can estimate the absolute position of a vehicle using a public online database of road signs. The estimated absolute position near a road sign is used to compensate the drift error of visual odometry (VO). In the first phase, the relative position between a road sign and a vehicle is estimated by matching a detected road sign image with the reference image from a public online database. Subsequently, the absolute position of the vehicle is calculated using the data from the database. Once the absolute position of the vehicle is estimated near a road sign, the current position of VO is updated to compensate the accumulated error. From a 24-km driving road test, it is validated that the proposed algorithm can estimate the absolute position of a vehicle within an error of 1.5 m. Moreover, a test of trajectory 3 km showed that it can maintain the drift error of VO within tens of meters. Our method is easy to be deployed, has low computation cost, and is accessible to a wide range of driving environments such as highways.