Visual Localization

A pose-graph map with visual landmarks is created from tracking local BRISK features on four fish-eye cameras mounted on the vehicle. Subsequent loop-closure detection, pose-graph relaxation and Bundle Adjustment generates a geometrically consistent map. Through offline localization of further datasets, the map can incorporate multiple sessions and cover a wide spectrum of appearance conditions. The resulting multi-session map can then be used for visual localization across weather and seasonal change in the long-term.

Leave a Reply