The 2nd International StepUP Competition for Biometric Footstep Recognition: From Steps to Strides

TL;DR — This is a report on the 2nd International StepUP Competition for pressure-based footstep biometrics, a specialized recognition benchmark that pushes toward robust, real-world identity verification from walking pressure data. It matters as a standardized evaluation driving progress in an underexplored biometric modality, though it fits none of the listed AI-advancement topics cleanly.

• Built on the StepUP-P150 dataset (200,000+ high-res dynamic footsteps from 150 people) plus an unreleased test set, targeting three challenges: generalization to unseen users with limited enrollment, robustness to footwear/speed domain shift, and left-right footstep fusion.
• New edition added extreme cross-domain conditions and moved from isolated footsteps to stride-level verification, enabling inter-step information fusion.
• Best result: 8.00% equal error rate (ArogyaPandit Research Team) using a spatiotemporal CNN plus ensemble-based scoring; inference-time normalization/calibration helped.
• Open problem remains: recognizing users in unseen personal footwear, especially with similar-looking distractors.