基于鱼眼相机和RGBD相机的异构协同SLAM

Heterogeneous collaborative SLAM based on fisheye and RGBD cameras

SLAM技术在卫星导航拒止环境下的自主导航中有着广泛的应用前景。结合单目鱼眼SLAM可获取更多纹理信息的优势和RGBD-SLAM可直接获取尺度信息的优势,设计基于单目鱼眼相机和RGBD相机的异构协同SLAM系统。首先设计特征点对三维灰度质心方向一致性检验方法以筛选异构图像之间的候选匹配点。然后设计异构图像之间的分步式光流-投影匹配方法,以实现鱼眼相机和RGBD相机之间高性能的特征点匹配与相对位姿估计。最后基于ORB-SLAM2框架,提出基于鱼眼相机和RGBD相机的异构协同SLAM系统框架。实验结果表明:相比于传统的特征点匹配方法,设计的特征匹配方法在异构相机的图像特征匹配任务中能够表现出更高的性能。相比于单目鱼眼SLAM系统和RGBD-SLAM系统,提出的异构协同SLAM系统在相机快速移动、相机贴近景物、低帧率、纹理缺失等条件下,以及在相机纯旋转运动、室外大场景等条件下性能更优,其鲁棒性、抗轨迹漂移能力和轨迹精度比单目鱼眼SLAM系统和RGBD-SLAM系统都有较大提升。

SLAM technology has a the potential of wide applications in autonomous navigation in satellite navigation denied environments.To combine the advantages of monocular fisheye SLAM to obtain more texture information and RGBD-SLAM to directly obtain scale information,a heterogeneous collaborative SLAM system is designed based on a monocular fisheye camera and a RGBD camera.Firstly,a method for checking the 3D gray centroid direction consis⁃tency between feature points is designed to screen the candidate matching points between heterogeneous images.Then,a step-by-step optical flow and projection matching method between heterogeneous images is designed to achieve high-performance feature point matching and relative pose estimation between fisheye and RGBD camera.Fi⁃nally,based on the ORB-SLAM2 framework,a heterogeneous collaborative SLAM system is proposed based on fish⁃eye and RGBD camera.The experimental results show that compared with traditional feature point matching meth⁃ods,the proposed feature point matching method shows higher performance in the task of image feature matching with heterogeneous cameras.Compared with the monocular fisheye SLAM and RGBD-SLAM system,the proposed heterogeneous collaborative SLAM system has better performance under the conditions of rapid camera movement,camera close to the scene,low frame rate,texture loss,pure rotation of the camera,outdoor large scenes,etc.,and demonstrates improved robustness,anti-drift ability and trajectory accuracy.