A Camera Center Estimation Based on Perspective One Point Method

In autonomous driving, estimating the position of an ego-vehicle and the distance between the vehicle and detected obstacles are very important. Visual-based methods have been used to estimate a camera pose equipped in the vehicle with successive images. Until now, Perspective-n-Point (PnP) based methods for estimating the position of a camera by calculating the world coordinates of three or more feature points and corresponding 3-dimensional object points have been used. This article proposes a novel camera center estimation method based on an Essential matrix and Perspective one Point (EP1P). This method employing epipolar geometry and geometrical constraints overcomes the limitation of the translation matrix, which is derived from the essential matrix to estimate camera center coordinates. To solve this problem, we performed coordinates formulation and derivation of equations for three consecutive images. To verify the accuracy of the proposed EP1P method and equations, simulations and experiments for the proposed method and several PnP-based methods have been performed. The results show that the proposed method has better performance than the existing methods. Since the proposed EP1P method needs only one feature point and a corresponding 3-dimensional object point, and the approach is different from the existing PnP-based methods with complex calculations, this would be a good alternative to estimate the vehicle position and surrounding objects’ position using the camera in limited situations.