New Algebraic Metric for Event Cameras Enables SOTIF-Compliant Safety Perception

A new paper from Arthur de Miranda Neto introduces a unified algebraic framework that adapts the Pearson Correlation Coefficient (PCC) for event camera streams, addressing a major gap in safety-critical perception for autonomous driving systems (ADS). Event cameras offer microsecond temporal resolution and 120-140 dB dynamic range, but existing quality metrics require a downstream task (e.g., detection accuracy) to assess stream integrity—a limitation incompatible with ISO 21448 (SOTIF) and ISO/PAS 8800:2024 certification standards. The BiasBench benchmark (CVPR 2025) explicitly flagged this gap.

Neto's framework lifts PCC to three standard event representations: Time Surface, Event Frame, and Voxel Grid, yielding three metrics: r-TS for stream integrity monitoring via ego-motion-predicted Time Surface comparison, r2-EF for adaptive ROI selection using only integer comparisons, and r-VG for temporal redundancy gating. The work establishes a structural isomorphism between the event camera's contrast-threshold mechanism (|ΔL| >= C) and the PCC-based change criterion. Demonstrations on a procedural-synthetic event stream include a tunnel-dip integrity-anomaly scenario where r_C drops from 0.93 (coherent flow) to below 0 (alarm). The paper lays theoretical foundations with empirical validation on real datasets reserved for follow-up work.