In the vision navigation system, the transformation from image coordinates to world coordinates is the essential part for the autonomous robot field based on crop rows detection using single camera. An approach of correcting the errors of the vision system is proposed in this paper, which can correct the errors of the target coordinates on the ground according to the theory of the lens distortion and coordinate transformations, and furthermore, the third-order B spline interpolation was used to improve the performance of the vision navigation system for the field autonomous robot. The experiment was implemented using the data obtained from the target board image, and the experimental results show that the approach proposed in this paper is effective.