|
|
|
| An RGB-D SLAM Algorithm for Robot Based on the Improved Geometric andMotion Constraints in Dynamic Environment |
| AI Qinglin, LIU Gangjiang, XU Qiaoning |
| Key Laboratory of Special Purpose Equipment and Advanced Manufacturing Technology, Ministry of Education & Zhejiang Province, Zhejiang University of Technology, Hangzhou 310023, China |
|
|
|
|
Abstract The pose estimation accuracy and robustness of the indoor robot using traditional visual SLAM (simultaneous localization and mapping) systems in dynamic environments are poor because the moving feature points are incorrectly incorporated into the camera pose calculation. In order to solve this problem, the feature points are divided into 5 categories, including static feature points, unknown feature points, suspected static feature points, dynamic feature points, and mismatched feature points. Static feature points are screened with strict geometric constraints. Unknown feature points are divided into suspected static feature points, dynamic feature points, and mismatched feature points using multi-frame observation information, and Kalman filtering is performed. Finally, the static feature points, suspected static feature points, and dynamic feature points are all used for pose optimization. The experiments on the TUM dataset in indoor environments with moving objects show that the overall performance (including accuracy, robustness, and running speed) of the proposed algorithm is superior to other dynamic SLAM algorithms.
|
|
Received: 29 April 2020
Published: 07 April 2021
|
|
|
|
|
|
|
|
|
