融合飞控传感器的无人机影像匹配方法

UAV Image Matching Method Fusing Flight Control Sensor

在拒止环境下无人机视觉定位问题中,ORB(Oriented FAST and Rotated BRIEF)具有高效、快速、稳定的特点,因而可以作为无人机实时影像与基准图匹配算法。本文针对传统ORB算法不具备尺度不变性,且在特征提取过程中存在特征点分布不均匀的问题,提出一种结合飞控传感器的无人机影像匹配方法。利用气压计与加速计经扩展卡尔曼滤波计算后的高度,通过双性线插值法定量改善无人机实时影像尺度,减少了ORB算法中构建影像金字塔所造成的系统开销,通过四叉树算法对提取的特征点均匀化,最后采用网格运动统计(Grid-based Motion Statistics, GMS)算法进行初次误匹配点剔除,再用随机抽样一致算法(RANSAC)筛选精确匹配点。实验结果表明,本文方法能够有效地减少无人机实时影像与基准图匹配时间,同时可以获得均匀的匹配点对,保持较高的匹配正确率。

In the problem of UAV visual positioning in the denial environment, ORB(Oriented Fast and Rotated Brief) can be used as real-time image matching algorithm for UAV due to its high efficiency, fast speed and stability. In view of the traditional algorithms of the ORB does not have scale invariance, and in the process of feature extraction has the problem of uneven distribution of feature points, a combination of multi-sensor UAV image matching method, the barometer and accelerometers through after the height of the extended Kalman filtering calculation, using the benchmark image scale and the proportion of the unmanned aerial vehicle flight level relations, through quantitative improve sex line interpolation UAV real-time image size scale, reduces the overhead caused by the build image pyramid. Quadtree algorithm was used to homogenize the extracted feature points. Finally, GMS algorithm was used to eliminate the initial mismatched points, and then Random Sampling Conformance Algorithm(RANSAC) was used to screen the exact matching points. Experimental results show that the proposed method can effectively reduce the matching time between UAV real-time image and reference image, obtain uniform matching point pairs, and maintain a high matching accuracy.