Abstract
To solve the problems of computational complexity and inaccuracy in classical vanishing point detection algorithms, such as the cascaded Hough transform, a vanishing point detection method based on constrained classification is proposed. First, the short line data are filtered to avoid interference in straight line detection, and then, the line segment is screened and classified by hierarchical clustering according to the image characteristics of the line segment and the variation pattern of angle similarity. Subsequently, Three types of straight line segments with the most significant angle differences are acquired. To prevent the optimization algorithm from getting stuck in the "wrong" local optimum neighborhood or failing to locate the global optimum, a set of constraints are set to further restrict the search. Afterward, the classified line segments are projected into a finite rhombic space, which are then quantified. The point with the maximum vote is eventually identified as the vanishing point. Experimental results show that the proposed method not only greatly reduces the computational complexity of vanishing points but also largely improves the accuracy of vanishing point detection.