Compensating drift of mono-visual odometry using road direction sign database

Mono-Visual odometry (Mono VO) is a widely implemented method to locate the ego-motion of a camera on a mobile platform. Since it is based on concatenating relative motions between image frames, it has a weakness of error accumulation that results erroneous drift for a long-range navigation. In this paper, we propose a drift compensation method for Mono VO by leveraging publicly available road sign database. In the first process of the proposed algorithm, it estimates the relative position of the camera on a vehicle with respect to the road sign using an online database that provides the positions and dimensions of the corresponding road signs. Then, the estimated positions are integrated in the Mono VO pipeline to update the current positions to compensate the accumulated drift. Unlike the previous research, the proposed solution neither uses a post-processing of optimization nor requires a loop closure, not to mention that the required database are easily available on online. Experiments are conducted to compare the proposed method with a Mono VO algorithm and results showed that the relative position drift is bounded within 0.5% for a 3km trajectory.