基于单目视觉的无人机障碍探测算法研究  

Obstacle detection for unmanned aerial vehicle based on monocular vision

作　　者：张跃东[1] 李丽[1] 刘晓波[1] 邢泽平[2] 

机构地区：[1]北京航空航天大学电子信息工程学院,北京100191 [2]北京航空航天大学仪器科学与光电工程学院,北京100191

出　　处：《激光与红外》2009年第6期673-676,共4页Laser & Infrared

基　　金：航空科学基金(No.20070151002);CAST创新基金(No.CAST200710)资助

摘　　要：针对低空飞行无人机的避障问题,研究了一种基于单目视觉的障碍物深度提取算法。在无人机前视摄像机获取的图像中,通过Harris角点检测算法提取角点作为特征点,使用归一化互相关算法进行角点的匹配,根据图像序列中特征点间距离的变化和无人机的运动计算障碍物的深度。对用于单个障碍物深度计算方法进行了改进,使用RANSAC算法区分不同深度的障碍物。仿真表明,该算法可以有效发现并区分不同深度的障碍物。For Unmanned Aerial Vehicles （UAV） flying at low altitude, a obstacle depth estimation algorithm based on monocular vision is discussed. Comer points are chosen as features using Harris Comer Detector from the images from a camera in front of the UAV. Comer points are matched using normalized correlation function. The depth of the obstacle is estimated by comparing the position change of image features with the position change of the UAV. For improving the depth estimation algorithm only can be used in single obstacle situation, RANSAC algorithm has been used to differentiate multiple obstacles at different depths. Simulation experiments show that this method is effective to detect and differentiate obstacles at different depths.

关 键 词：单目视觉 图像处理 无人机 障碍探测 RANSAC算法 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]