同时场景构建和定位算法是机器人自主导航的重要组成部分。针对传统算法不能应用于室外环境和缺乏定量分析的缺点,提出了一种单摄像机恢复场景三维结构和摄像机位姿的新算法。提出了视频序列关键帧提取方法,降低了运算复杂度; 利用特征点对和摄像机内参量计算场景三维结构和关键帧的位姿并提出一种估计关键帧位姿的简便方法; 最后,提出一种兼顾优化效果和运算复杂度的自适应光束法平差算法优化场景结构和摄像机位姿,并生成适于机器人导航的数字高程图。室内和室外多种场景下的定量和定性实验结果表明,绕行误差低于4%,该算法能够接近实时准确实现同时场景构建和摄像机定位。

Simultaneous localization and mapping (SLAM) is one of the most important components in robot navigation. A novel SLAM algorithm based on monocular vision is proposed to overcome the difficulties in outdoor applications and quantitative analysis with traditional methods. Firstly, a key frame selection method is proposed to reduce the computational cost. Then the three-dimensional (3-D) structure of the environment and the positions of the camcorder are estimated based on matched feature points and the intrinsic parameters of the camcorder. A simple method with reasonable optimizing effect and computing cost is applied to get position and orientation of the camcorder. Finally, an adaptive bundle adjustment is adopted to optimize the 3D structure of the environment and the positions of the camcorder simultaneously. Digital elevation map (DEM) which is more suitable for robot navigation is also obtained. Quantitative and qualitative experimental results show that the loop closure error is less than 4%. The algorithm can reconstruct the environment and localize the camcorder accurately in nearly real time.