LC-Flow: Learning Local Continuous Optical Flow and Confidence from events

2026-05-23

AI Summary

Existing event-based optical flow methods struggle to simultaneously achieve microsecond-level temporal continuity and leverage historical context: frame-based approaches introduce latency and are prone to overfitting, while stateless local models suffer from limited accuracy due to their neglect of temporal information. This work proposes LC-Flow, the first learning-based, purely local, continuous, and stateful event-driven optical flow estimator. LC-Flow employs a continuous local recurrent network over a spatial grid to incrementally maintain hidden states as events arrive, jointly predicting optical flow and a learnable confidence measure. This confidence effectively resolves local ambiguities and enables downstream filtering and multi-scale confidence-guided aggregation to reconstruct globally consistent flow. LC-Flow achieves state-of-the-art performance among local methods on MVSEC and DSEC, and its confidence-guided aggregation surpasses heavyweight frame-based methods that rely on global priors, setting a new overall state of the art on MVSEC.

Abstract

Event cameras capture brightness changes asynchronously with microsecond resolution, yet existing optical flow methods fail to fully exploit this temporal continuity. Frame-based approaches impose artificial accumulation latency and suffer from domain overfitting, while model-based local methods operate statelessly, discarding temporal history between predictions and yielding inaccurate flows. We propose \textbf{LC-Flow}, the first temporally continuous, learning-based optical flow estimator that operates purely from local events. At its core, a Continuous Local Recurrent Network maintains persistent hidden states per spatial grid, incrementally accumulating temporal context as events arrive. Unlike frame-based methods constrained to fixed accumulation windows, and unlike stateless model-based methods that recompute motion from scratch at each step, LC-Flow produces sparse local flow estimates at arbitrary timestamps with full motion history. To address the inherent ambiguity of local observations, we jointly learn a confidence score that quantifies the reliability of each prediction, explicitly handling event sparsity and the aperture problem. This confidence serves a dual role: filtering unreliable estimates for downstream tasks such as visual odometry, and providing principled weights for a multi-scale confidence-guided aggregation that reconstructs globally consistent flow from the sparse local outputs. LC-Flow achieves state-of-the-art performance among local methods on both MVSEC and DSEC, while the confidence-guided aggregation establishes a new overall state-of-the-art on the MVSEC benchmark, surpassing heavy frame-based networks that rely on global spatial priors.

Problem

• event camera
• optical flow
• temporal continuity
• local estimation
• aperture problem

Innovation

• event camera
• optical flow
• temporal continuity
• confidence estimation
• local recurrent network