Unaligned but Safe -- Formally Compensating Performance Limitations for Imprecise 2D Object Detection
This addresses safety concerns in autonomous systems by providing formal guarantees for object detection, though it is incremental as it builds on existing detection limitations.
The paper tackles the problem of imperfect alignment in 2D object detection by formally proving a minimum bounding box enlargement factor to cover ground truth, and shows it can be reduced with a motion planner buffer, connecting statistical and worst-case evidence.
In this paper, we consider the imperfection within machine learning-based 2D object detection and its impact on safety. We address a special sub-type of performance limitations: the prediction bounding box cannot be perfectly aligned with the ground truth, but the computed Intersection-over-Union metric is always larger than a given threshold. Under such type of performance limitation, we formally prove the minimum required bounding box enlargement factor to cover the ground truth. We then demonstrate that the factor can be mathematically adjusted to a smaller value, provided that the motion planner takes a fixed-length buffer in making its decisions. Finally, observing the difference between an empirically measured enlargement factor and our formally derived worst-case enlargement factor offers an interesting connection between the quantitative evidence (demonstrated by statistics) and the qualitative evidence (demonstrated by worst-case analysis).