Unaligned but Safe -- Formally Compensating Performance Limitations for Imprecise 2D Object Detection

TL;DR

This paper addresses safety concerns in 2D object detection by formally analyzing and compensating for bounding box misalignments, ensuring reliable performance within safety-critical applications.

Contribution

It provides a formal proof of the minimum bounding box enlargement needed for safety and proposes a method to adjust this factor based on motion planning buffers.

Findings

Derived the worst-case bounding box enlargement factor mathematically.

Connected empirical measurements with formal worst-case analysis.

Showed how to adjust detection thresholds for safety in motion planning.

Abstract

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).