一种基于类间方差的地平线检测算法

 针对视觉导航中用于微型飞行器（MAV）稳定飞行控制的地平线提取问题,提出一种基于类间方差的地平线检测算法。该算法使用角度和距离两个变量进行图像中直线的穷举,选取像素的蓝色分量作为区分天地的特征,利用类间方差构造判别准则进行直线的最优选择,从而实现地平线的检测。实验结果表明：与基于色彩协方差矩阵的地平线检测算法相比,本文算法对地平线检测的平均角度误差减少1°,距离误差减少3个像素,角度和距离参数的检测正确率分别提高1.98%和4.95%,且具有良好的速度特性。

A Horizon Detection Algorithm Based on Between-class Variance Analysis

To detect the horizon for the steady flight control of micro air vehicles (MAV) in vision navigation, a horizon detection algorithm based on between-class variance analysis is proposed. The angle and distance are used as parameters to exhaust the straight lines in the image. The blue color component of the pixel is chosen as the feature for distinguishing the sky and the ground. A criterion based on the between-class variance is set up for determining whether or not a straight line is the horizon within the exhaustive straight lines. Experimental results show that, compared with the horizon detection algorithm based on color covariance matrix calculation, the proposed algorithm performs better. The average detection error decreases by 1 degree in angle and 3 pixels in distance. The correct detection rate increases by 1.98% in angle and 4.95% in distance at a satisfactory speed.