GeoIMO: Geometry-Driven Independent Motion Classification for Event Cameras

2026-06-23

AI Summary

This work addresses the limitation of existing automotive event datasets, which rely on frame-level appearance annotations and are thus ill-suited for motion-aware tasks. The authors propose a geometrically driven, annotation-free, and learning-free method that leverages the inherent ego-motion structure within event streams to distinguish static background from independently moving objects. By introducing a yaw-compensated expanding focus model to estimate global background motion, and combining it with a scale-invariant residual metric and a temporal stability mechanism to assess local motion deviations, the approach significantly enhances robustness and adaptability in complex driving scenarios—particularly during vehicle turns. Evaluated on the MVSEC and Prophesee 1 Megapixel Automotive Detection datasets, the method demonstrates consistent performance while achieving a favorable trade-off between accuracy and computational efficiency.

Abstract

Existing automotive event datasets rely on appearance-based annotations from frame pipelines, making them poorly suited for motion-aware event perception. We present a geometry-driven, annotation-free framework that classifies detected objects as static or independently moving by exploiting ego-motion structure directly from the event stream. A Focus of Expansion model with yaw compensation estimates global background motion, while objects are labeled as moving when local motion deviates from this prediction, as quantified by a scale-invariant residual. Temporal stabilization improves robustness across consecutive event windows. The method requires no learning, no manual motion labels, and works with any input bounding boxes. Experiments on MVSEC and the Prophesee 1 Megapixel Automotive Detection dataset demonstrate consistent performance across diverse driving scenarios, with yaw compensation improving results during turns and a simple translational local model offering a favorable accuracy-efficiency trade-off.

Problem

• event cameras
• independent motion classification
• ego-motion
• annotation-free
• motion perception

Innovation

• event cameras
• geometry-driven
• independent motion classification
• Focus of Expansion
• annotation-free