A Vector Map Fusion Method Based on Cylindrical Point-Cloud Representation Combined with a Spatio-Temporal Attention Mechanism [Patent]

This invention relates to the technical field of autonomous driving and specifically provides a vector map fusion method, computer device, and computer-readable storage medium, aiming to solve the problem of improving the accuracy of fusing multiple batches of homologous vector maps. To this end, the method provided by this invention includes: converting multiple batches of homologous 3D vector maps into 2D bird’s-eye-view map images; obtaining the element features of target map elements in the map images based on an attention mechanism; obtaining the element information of the target map elements according to the element features; and obtaining the fusion result of the multiple batches of 3D vector maps according to the element information. This method achieves end-to-end 3D vector map fusion without manually designed rules, effectively avoiding the rule-based fusion of the prior art, which tends to cause problems such as broken lines and disordered lines at the fusion points, thereby improving fusion accuracy.

Download paper here