无人飞行器视觉SLAM特征检测和匹配算法研究

Research on Feature Detection and Matching Algorithm of Unmanned Aerial Vehicles Visual SLAM

摘要:

近几年图像局部特征检测和描述在机器人视觉中得到了广泛的应用,鲁棒的、快速且高精度的视觉特征检测和描述算法对飞行器进行实时的位姿估计和地图构建具有决定性意义；针对四旋翼无人飞行器平台的RGB-D传感器同时定位与地图构建(SLAM),讨论FAST、STAR、SIFT和SURF等检测算法和ORB、FREAK和SURF等匹配描述符的性能,对不同的特征算法进行对比评估出最合适的特征检测方法和匹配描述符；最后,基于Eclipse与OpenCV平台进行了实验,实验结果表明FAST检测和FREAK描述符比其他方法更适用于四旋翼飞行器在板视觉SLAM,且能基本满足实时性。

Abstract:

Recently,image local feature detection and description has been widely used in robot vision,It is decisive significance of a robust,rapid and high accuracy of visual feature detection and description algorithm for unmanned aerial vehicles real-time pose estimation and mapping. In view of the unmanned aerial vehicles RGB-D sensor Simultaneous Localization and Mapping(SLAM) characteristics,this paper discussed the performance of FAST,STAR,SIFT,SURF detection algorithm and the ORB,FREAK,SURF descriptor,and then compared different algorithm to find the most suitable feature detection method and the descriptor. Finally,the results of the experiment which is based on Eclipse and OpenCV platform,show that the FAST detection and FREAK descriptor is better than other methods suitable for unmanned aerial vehicles real-time visual SLAM.