Fast and robust RGB-D relocalisation
In the demo, we map a room about 16 square metres in real-time and then relocalise the poses of lost RGB-D sensor several times. We deliberately kidnap our sensor to unknown positions which are far away from the path on which we build the map. Overall the system is running at 15+ frames per seconds. The dense relocalisation and mapping approach is based on the combination of [1] and [2]. For more details including source codes and documents, please check out our webpage at lishuda.wordpress.com [1] Li, S., & Calway, A. (2015). RGBD Relocalisation Using Pairwise Geometry and Concise Key Point Sets. In ICRA. [2] Li, S., & Calway, A. (2016). Absolute pose estimation using multiple forms of correspondences from RGB-D frames. In ICRA (pp. 4756–4761).
In the demo, we map a room about 16 square metres in real-time and then relocalise the poses of lost RGB-D sensor several times. We deliberately kidnap our sensor to unknown positions which are far away from the path on which we build the map. Overall the system is running at 15+ frames per seconds. The dense relocalisation and mapping approach is based on the combination of [1] and [2]. For more details including source codes and documents, please check out our webpage at lishuda.wordpress.com [1] Li, S., & Calway, A. (2015). RGBD Relocalisation Using Pairwise Geometry and Concise Key Point Sets. In ICRA. [2] Li, S., & Calway, A. (2016). Absolute pose estimation using multiple forms of correspondences from RGB-D frames. In ICRA (pp. 4756–4761).