结合视觉里程计的微小型空中机器人SLAM研究

The SLAM combining with visual odometry for small-scale aerial robots

经典的基于"平滑摄像机模型"的单目视觉同步定位与地图构建方法无法适用于具有复杂飞行模式的微小型空中机器人。针对这个问题,提出一种结合视觉里程计的单目视觉同步定位与地图构建方法。该方法通过视觉里程计直接估计机器人机载摄像机相对位姿变化,并将这些位姿信息嵌入基于EKF的单目视觉同步定位与地图构建算法中。同时,在采用视觉里程计进行位姿估计时,针对可能出现的退化问题,采用特征分类的策略,提高了估计的鲁棒性。将方法应用于一套真实的微小型智能无人直升机系统上,实验数据验证了方法具有良好的适用性和实用性。

The standard monocular simultaneous localization and mapping （SLAM）approach based on the smooth moving camera model cannot be applied to the small-scale aerial robots having many complex flight styles. This paper proposes a novel monocular SLAM method for a small-scale aerial robot. The relative pose information of the moving camera are received by using visual odometry, and a novel strategy based on classifying features is applied to deal with nearly degenerate situation that often arises in practical applications. Then the estimated relative poses are integrated into the extended kalman filter （EKF） based SLAM algorithm. The experiment results show good compatibility and practicality.