摘要
针对传统惯性导航累积误差大的缺陷,研究提出了一种视觉导航姿态估计方法;首先提取图像的局部特征,分别对SURF(speeded up robust features)、SIFT(scale-invariant feature transform)及GPU-SIFT特征提取算法进行了比较;保证算法精度及实时性后,将实时图与基准图库进行局部特征匹配,并利用EPnP算法进行飞行器的六自由度参数解算;实验结果表明GPU-SIFT算法精度最高,且随着图像分辨率的提高,其计算速度相比SURF和SIFT算法有了显著提高,该方法在一定条件下具有较高的位姿精度和良好的实时性。
A method of pose estimation in vision--based navigation is presented aiming at the defects of cumulative error in traditional inertial navigation. The first step is to extract the local features and it is carried out respectively by SURF, SIFT and GPU--SIFT. With real --time image and reference image matched, the position and attitude is calculated according to the EPnP algorithm. The result shows that the GPU--SIFT algorithm gets the highest accuracy. As the image resolution improved, the computing speed of GPU--SIFT shows remarkable superiority than SURF and SIFT. The simulation shows that the method works well in pose estimating.
出处
《计算机测量与控制》
2015年第4期1371-1374,共4页
Computer Measurement &Control
关键词
特征匹配
视觉导航
姿态估计
feature detection
vision--based navigation
attitude estimation