基于自适应直线拟合的角点检测

角点检测是计算机视觉的一个基本问题，进行角点检测的关键是估算曲率。笔者给出了一种新的基于参考点的前后曲线方向估计的曲率计算方法。我们以离散点远离直线垂直距离误差最小为目标进行直线拟合，同时根据误差大小来自适应的选择拟合窗口的大小。很好地解决了角点曲率计算时拟合窗口大小和计算精度间的矛盾，实验证明，该方法抗干扰性好，且运算量不大，对于存在明显转折点的曲线角点有很好的检测和定位能力。

Corner Point Detection Based on Adaptive Line Approximation

The detection of corner points is one of the basic problems of computer vision. Usually the key question of detecting corner points is how to calculate curvature. This paper presents a novel approach of calculating the backward and forward direction of reference point based on the adaptive line approximation. Our line approximation method objects minimize the distance error and the length of approximation widow adaptively changes with the approximation error. Our adaptive method finds a compromise between the length of approximation widow and the calculation precise. The experiments results show, the method can robustly, effectively and precisely locate corner point for curves with apparent corners with little computation.