Vehicle position estimation using geometric constants in traffic scene

Abstract

Determination of the correct positional relation is vital for human driver. For unmanned vehicles, the obstacle position in front of the view is also necessary for collision detection. The paper is devoted to the problem that estimates the vehicle position in real world using only a single 2D image. The estimation is an ill-posed problem due to the projective transform; however, through incorporating the geometric constants in the traffic scene, we proposed a solution that calculates the position with sufficient accuracy. The contribution of the proposed method is the use of two geometric constants: the standard size of license plate and the lane-width. Observation shows that the size of license plates has limited patterns and lane-width between two adjacent lanes is usually constant. Introduction of the two constants compensates the uncertainty caused by lack of depth in the mapping between the detected license plates and its position in real world. The conducted experiments show that compared to the conventional methods, the proposed one is accurate for estimating the position.