Research on dynamic scene visual SLAM based on deep learning
To address the issues of low accuracy and poor robustness of visual SLAM systems in dynamic environments,a deep learning based dynamic scene visual SLAM method was proposed.Firstly,using improved YOLOv5 algorithm and optical flow con-straint algorithm to remove feature points on dynamic targets.Only retain static feature points that do not affect the construction of the map.Then,the GMS-RANSAC fusion algorithm was used to eliminate mismatches and accurately estimate the camera pose.Finally,by integrating word bags and YOLOv5 object detection algorithm,the efficiency and accuracy of loop detection have been improved.By conducting experiments using the highly dynamic walking series in the TUM dataset,the experimental results show that the proposed method can maintain stable robustness and accuracy in dynamic environments.
万方数据
维普
