Visual Positioning Fusion Pose Candidate Picture Screening

The patent segments the screening process into two stages: first using GPS-based spatial distance filtering to narrow down candidate frames from the large database, then applying scenario similarity screening only to this reduced subset. This segmentation avoids the inefficiency of applying scenario similarity screening to all frames in the large database, thus resolving the contradiction between handling large-scale databases and maintaining screening efficiency.

The patent merges GPS-based spatial distance filtering with scenario similarity screening to create a hybrid recall mechanism. The spatial distance filter uses GPS coordinates to pre-select candidate frames within a certain distance range, while scenario similarity screening further refines the selection. This combination compensates for the weaknesses of each individual method: spatial filtering handles illumination changes and viewing angle differences better, while scenario similarity provides semantic understanding. The merged approach maintains reasonable complexity while significantly improving recall accuracy under challenging conditions.

The patent applies GPS-based spatial distance filtering as a preliminary action before the main scenario similarity screening process. By first using the readily available GPS coordinates to filter out frames that are spatially too distant, the system reduces the number of frames that need to undergo computationally intensive scenario similarity analysis. This preliminary filtering action significantly reduces the overall screening workload while the complexity added is minimal, as GPS coordinate comparison is a simple mathematical operation.