基于语义分割动态特征点剔除的SLAM算法

SLAM algorithm based on semantic segmentation and dynamic feature point elimination

针对动态物体容易干扰SLAM建图准确性的问题,提出了一种新的动态环境下的RGB-D SLAM框架,将深度学习中的神经网络与运动信息相结合。首先,算法使用Mask R-CNN网络检测可能生成动态对象掩模的潜在运动对象。其次,算法将光流方法和Mask R-CNN相结合进行全动态特征点的剔除。最后在TUM RGB-D数据集下的实验结果表明,该方法可以提高SLAM系统在动态环境下的位姿估计精度,比现有的ORB-SLAM2的表现效果更好。

Aiming at the problem that dynamic objects tend to interfere with the accuracy of SLAM mapping,this paper proposed a new RGB-D SLAM framework for dynamic environments,which combined neural networks in deep learning with motion information.Firstly,the algorithm used the Mask R-CNN network to detect potential moving objects that might generate dynamic object masks.Secondly,the algorithm combined the optical flow method and Mask R-CNN to remove full dynamic feature points.Finally,the experimental results under the TUM RGB-D dataset show that this algorithm can improve the pose estimation accuracy of the SLAM system in dynamic environments and perform better than the existing ORB-SLAM2.