Sliding Windowed Optimization Algorithm for Fusion of Redundant Stereo Visual Odometries

• Elizabeth V. Cabrera-Avila UFRN http://orcid.org/0000-0002-8578-4515
• Bruno M. F. Silva UFRN https://orcid.org/0000-0002-7780-7254
• Luiz M. G. Gonçalves UFRN https://orcid.org/0000-0002-7735-5630

Resumo

We propose a sliding windowed graph-optimization based approach for the fusion of two stereo visual odometries with 6 degrees of freedom poses. Stereo odometries are calculated independently, using pairs of stereo images that are captured by stereo cameras mounted on the top of a moving platform. We implemented our sliding windowed approach using the public library LIBVISO2. Our results are compared against the two recognized SLAM frameworks ORB-SLAM2 and UCOSLAM. The relative pose error of the fused poses decreases by up to 94% in relation to the error of single stereo odometry and by up to 91% compared with the results of UCOSLAM.

Palavras-chave: Visualization, Simultaneous localization and mapping, Motion estimation, Robot vision systems, Redundancy, Cameras, Libraries, Sensor fusion, Stereo Visual Odometry, Sliding Windowed Optimization, Errors Accuracy