Year: 2023
Type: conference-paper
Abstract: With the aging of society's population and the outbreak of new coronary epidemics, there is an increasing demand for human-machine interactive intelligent mobile robots. Efficient and accurate semantic maps are the guarantee for the operation of intelligent mobile robots, in which Simultaneous Localization And Mapping (SLAM), as a key technology, has become a current research hotspot. However, in indoor scenes, traditional SLAM schemes build maps with fewer environmental features and rely on the extraction and matching of feature points, and their mapping performance degrades for scenes with lack of texture or dynamic changes. In response to the above problems and current situation, this paper studies the visual SLAM algorithm in indoor scenes and proposes an improved ORB-SLAM2 algorithm, which addresses the problems that the original algorithm has few weak texture region features, is greatly affected by dynamic targets and has low performance of loopback model, proposes a weak texture region feature enhancement strategy and optimizes the feature point distribution; introduces a dynamic feature detection mechanism to reject dynamic target features; optimizes the closed-loop detection model is optimized, which effectively improves the performance of indoor mapping. The experiments on TUM dataset show that the accuracy of ATE index is improved by 13.19% and the accuracy of RPE index is improved by 12.57% with the combined effect of the three improvement strategies in this paper.
Authors: Jin Wang, Xiaobo Sun, Mingze Wang, Yufei Xie, Ze Wang
Institution: Wuhan University of Technology
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: bronze
OpenAlex ID: W4385299265