Abstract: The existing vanishing point detection methods mostly remove outliers by statistical analysis of the vanishing point's candidates, and do not make full use of the straight line's parameter information, which leads to low precision and large calculation. In the paper, a robust vanishing points detection method with line parameter information was proposed. Firstly, the algorithm extracted and analyzed the line parameter information at Manhattan direction, and proved them with linear relation. Secondly, the parameters' linear model was established with robust regression algorithm, and then the outliers were removed to get effective lines. Finally, it estimated the optimal vanishing point at Manhattan direction from the obtained effective lines. The experimental results show that the average error of the focal length, which is calibrated by the vanishing points detection algorithm, is 1.05 pixel. Therefore, the detected vanishing points can be effectively applied to the camera calibration.

Key words: scene understanding, vanishing point detection, Manhattan direction, robust regression

摘要： 针对现有的灭点检测方法未充分利用产生灭点的直线的参数信息，导致检测精度较低、计算量较大等问题，提出了一种利用直线参数信息的稳健灭点检测算法。首先采用Canny算子和Hough变换相结合的方法提取出建筑物图像中较长的稀疏直线，通过分析直线的参数信息，对不同方向直线进行聚类，并证明了各方向的直线参数满足线性分布关系;然后利用稳健回归算法建立直线参数的线性模型，并据此去除外点，获得产生有效候选灭点的有效直线束;最后根据有效直线束计算曼哈顿方向的最优灭点。实验结果表明，所提的灭点检测算法应用于规则建筑物图像的摄像机标定时，焦距的平均误差为1.05像素。

关键词: 场景理解, 灭点检测, 曼哈顿方向, 稳健回归