基于特征点法的自适应SLAM改进方法

An improved method of adaptive SLAM based on ORB-SLAM2

由于传统的机器人定位导航的方法多为特征点法,由于特征点法需要在较为丰富的纹理环境下进行特征点的提取,故特征点法易在低纹理环境下失效。针对机器人定位导航在低纹理环境下失效的问题,提出一种机器人定位导航改进方法,通过对于相机图像的预处理,分选出低纹理环境图片,并把分选好的图片流入不同的系统进行处理,最后结合得到的位姿信息与惯性测量单元(IMU)中的数据进行融合,得到一个可以适用于低纹理环境下的机器人定位导航系统。使用EuRoC数据集对于改进的机器人定位导航系统进行验证。实验结果表明,该系统可以识别ORB-SLAM2无法正常运行的低纹理环境,误差比LSD-SLAM的减小70%。

Due to the traditional robot positioning and navigation is the most common feature point method, and the feature point method needs a rich texture environment to extract the feature points, so the feature point method is easy to fail in the low texture environment.In order to solve the problem that robot positioning and navigation fails in low-texture environment, this paper proposes an improved method for robot positioning and navigation.Through the preprocessing of camera images, low-texture environment pictures are sorted, and the sorted pictures are sent to different system performs processing, and fused infonmation mix up the obtained pose information, which the data is in the inertial measurement. The mixed infrmation help system is suitable for low-texture environment. The improved robot positioning and navigation system is validated using the EuRoC dataset.The experimental results show that the system can identify low-texture environments where ORB-SLAM2 cannot operate normally, and the error is 70% smaller than that of large scale direct monocular SLAM(LSD-SLAM).