Feature Matching Based on Target Detection Optimization in Dynamic Scene

Abstract

To improve the accuracy of ORB_SLAM2 pose estimation in dynamic environment, our paper proposes an effective method which uses the target detection algorithm to select the ROI of the key frame of the image, and then performs feature point matching. This method can be divided into three steps. Firstly, targets are detected and the ORB feature points are extracted from the input image, the corresponding pixel position coordinates of the object are eventually obtained in the scene; Secondly, feature point matching are performed on these ROI regions; Thirdly, the feature points with the relatively stable regions of interest are used to perform pose estimation. To verify the effectiveness of our method, the proposed method is tested on the public dataset. The experimental results show that the improved method can not only shorten the matching time, but also greatly improve the accuracy when compared with the traditional method. Thus, the original ORB_SLAM2 system can significantly improve the accuracy of pose estimation in dynamic environment.