Abstract: A vision-and-laser-fused 2.5D map building method includes: calculating inter-image frame transformation according to an RGB-D image sequence to establish a visual front-end odometer; taking a visual front-end initial estimation as an initial value of scanning matching, and performing laser front-end coarse-grained and fine-grained searches; performing loop closure detection, and performing back-end global optimization on a 2.5D map according to a detected closed loop; and performing incremental update on visual feature dimensions of the 2.5D map, and performing occupation probability update on grid dimensions. The 2.5D map is built using a method of fusing a laser grid and visual features, and compared with a pure laser map and a pure visual map, richness of dimensions is improved, and completeness of information expression is improved; the 2.5D map building method is not influenced by single sensor failure, and can still stably work in a scenario of sensor degradation.

Abstract: A vision-and-laser-fused 2.5D map building method includes: calculating inter-image frame transformation according to an RGB-D image sequence to establish a visual front-end odometer; taking a visual front-end initial estimation as an initial value of scanning matching, and performing laser front-end coarse-grained and fine-grained searches; performing loop closure detection, and performing back-end global optimization on a 2.5D map according to a detected closed loop; and performing incremental update on visual feature dimensions of the 2.5D map, and performing occupation probability update on grid dimensions. The 2.5D map is built using a method of fusing a laser grid and visual features, and compared with a pure laser map and a pure visual map, richness of dimensions is improved, and completeness of information expression is improved; the 2.5D map building method is not influenced by single sensor failure, and can still stably work in a scenario of sensor degradation.