Details

Autor(en) / Beteiligte

• Liu, Xuan-ang

Titel

A New Algorithm for Eliminating Mismatch

Ist Teil von

• Journal of physics. Conference series, 2023-08, Vol.2577 (1), p.12002

Ort / Verlag

Bristol: IOP Publishing

Erscheinungsjahr

2023

Quelle

Free E-Journal (出版社公開部分のみ）

Beschreibungen/Notizen

• Abstract Aiming at the problems of mismatching and low matching efficiency in the process of monocular initialization of visual SLAM, this paper proposes an improved initialization method. Based on the ORB-SLAM2 algorithm, the concept of probability statistics is introduced in its monocular initialization process. By adding a moving grid before the RANSAC algorithm eliminates the mismatching, the region where the feature matching is most likely to be true is found, and the mismatching is eliminated in this region. Finally, a geometric constraint is carried out to retain the correct matching to the greatest extent, thus speeding up the feature matching and retaining more interior points. In this paper, the improved ORB-SLAM2 algorithm is verified in the TUM data set. The experimental results show that the initialization time of the improved algorithm is reduced and the trajectory accuracy is improved. Applying the improved algorithm to the AGV car has good accuracy and robustness in the real factory environment.