One-Step Event-Driven High-Speed Autofocus

TL;DR

The paper tackles the challenge of high-speed autofocus in dynamic scenes by introducing the Event Laplacian Product (ELP), a novel focus detection function that fuses event data with grayscale texture to enable one-step AF. By deriving a theoretical link between temporal changes and spatial Laplacian cues, and by incorporating an adaptive filter, ELP produces a sharp sign mutation at the focus position, allowing real-time cessation of focus motion without iterative searches. Empirical results on synthetic and real datasets (DAVIS346 and EVK4) show that ELP reduces focusing time by up to two-thirds and lowers focusing error by up to 24x (DAVIS346) and 22x (EVK4) compared to state-of-the-art event-driven methods, while supporting an event-only pipeline via EvTemMap. The approach robustly handles motion and low-light conditions, offering a practical, one-step AF solution for event cameras with broad applicability in robotics and vision systems.

Abstract

High-speed autofocus in extreme scenes remains a significant challenge. Traditional methods rely on repeated sampling around the focus position, resulting in ``focus hunting''. Event-driven methods have advanced focusing speed and improved performance in low-light conditions; however, current approaches still require at least one lengthy round of ``focus hunting'', involving the collection of a complete focus stack. We introduce the Event Laplacian Product (ELP) focus detection function, which combines event data with grayscale Laplacian information, redefining focus search as a detection task. This innovation enables the first one-step event-driven autofocus, cutting focusing time by up to two-thirds and reducing focusing error by 24 times on the DAVIS346 dataset and 22 times on the EVK4 dataset. Additionally, we present an autofocus pipeline tailored for event-only cameras, achieving accurate results across a range of challenging motion and lighting conditions. All datasets and code will be made publicly available.

Figures (10)

• Figure 1: Visualization of ELP, event frames, images and Laplacian at different $\Delta v$.
• Figure 2: Event-driven one-step AF setup and pipeline. (a) Hardware setup. (b) Collected real events. (c) The pipeline of grayscale image acquisition and ELP value calculation.
• Figure 3: The pipeline of PSF-based focus event simulator.
• Figure 4: Visualization of a violent shake scene. Top row: temporal variation of ELP and EPR across the entire focus stack. Bottom row: grayscale images at different $\Delta v$ overlaid with focus events.
• Figure 5: Visualization of two scenes from the DAVIS dataset. Top row: temporal variation of ELP and EPR across the entire focus stack. Bottom row: grayscale images corresponding to the focus positions determined by the three event-driven methods, alongside the GT. Green boxes indicate the focus ROI.