In this paper, inspired by RANSAC and Hough Transform voting procedure, a modified RANSAC algorithm of detection of cylindrical reference target using LRF for positioning system is proposed. Utilizing the orientation-invariant property of the circle, cylindrical shaped bar is chosen as the reference target of the positioning system. The given radius of the cylinder and convex arc shaped contour observed by LRF are two important constraints of the model of circle used in this positioning system. Simulation results prove that under given conditions the proposed circle detection algorithm works better than original RANSAC when occlusion and noise make the ratio of qualified inliers decrease. The reason is that a voting procedure can reduce the effect of noise and guarantee a more reliable model hypothesis. After extraction of enough inliers by proposed circle detection algorithm, maximum likelihood estimation is implemented to estimate the precise center position of the reference target.