Online scene change detection is the task of noticing what has changed in an environment that was previously mapped, while still on site. The agent must work from unconstrained viewpoints and cannot rely on future observations, making the problem inherently causal.
The paper introduces O-SCD++, a method that addresses this setting through recursive evidence fusion. By combining evidence as it arrives, the approach avoids replay—meaning it does not need to revisit or reprocess past data—while still detecting changes in real time.
The abstract is brief, but the contribution appears to be a principled way to maintain an up-to-date scene representation under causal constraints. Further details on evaluation and comparison with prior methods are not included in the available text.