Abstract: The traditional method can only calculate the orthogonal vanishing point of the calibration
image, dose not take into account the error of straight line detection, the length of the straight line and
the location relationship between the candidate vanishing point and the constrained line has the
impact on the accuracy of the vanishing point detection. For such problems, this paper proposes an
un-calibrated single visual image of orthogonal vanishing point detection algorithm. First, use
J-Linkage algorithm to complete the initialization of the vanishing point estimates, and get the set of
assumed vanishing points. Second, according to the consistency between the assumed vanishing
points and lines, length of line segment to get the exact vertical vanishing point based on a voting
mechanism. Third, calculate the parameters of the image camera by the nature and properties of the
vanishing point and vanishing line. Then based of the orthogonal characteristics of the vanishing point
to get the exact horizontal and depth direction of the vanishing point. By the introduction of a new
consistency measurement method between the assumed vanishing point and the line of image, this
method not only gets rid of the dependence on the camera parameters, but also reduces the impacts of
the impacts of errors of line segment extraction, length of line segment as well as the relative position of
vanishing point and the constrained line on the precision of vanishing points detection, in the case of unknown image camera parameters this method can get accurate information about the three orthogonal
vanishing point. This method this thesis applied can achieve better results in indoor scenes.

Key words: single visual image, un-calibrated, voting, orthogonal vanishing points, N-vectors