扩展Kalman滤波框架下的地平线检测算法

Horizon detection algorithm in extended Kalman filter framework

中文关键词  地平线检测  边缘检测  暗原色先验  无人机  姿态估计

英文关键词  horizon detection  edge extraction  dark channel prior  unmanned aerial vehicle  attitude estimation

中文摘要

      针对基于视觉的无人机姿态测量中的地平线检测问题, 提出一种结合图像处理和扩展Kalman滤波的地平线检测算法. 图像处理部分先从图像边缘中提取直线来获取候选地平线, 再根据图像暗原色先验来选取地平线. 扩展Kalman滤波部分将地平线观测模型与机身旋转模型结合起来, 获得观测值和残差协方差的先验信息, 用于判别检测到的地平线是否正确, 并在检测到错误地平线时进行修正. 实验结果表明, 该算法能够有效地检测地平线, 对复杂场景具有鲁棒性.

英文摘要

      To detect the horizon in the pose measurement for a vision-based unmanned aerial vehicle (UAV), we propose a horizon-extraction algorithm which combines the image processing and extended Kalman filter (EKF). The image processing part extracts straight lines from the image edges as candidates of horizon; and then, it uses the dark channel prior of the image to decide the horizon. The EKF part of the algorithm combines a horizon line observation model with a fuselage rotation model to obtain the predicted observation and innovation covariance, which are employed for judging whether a correct horizon is detected. Correction will be carried out when wrong horizon is detected. The experimental results show that the proposal algorithm can effectively detect the horizon, and is robust to complex scenes.