High-frequency scanning and depth estimation in event-based structured-light system

Timeline:

Using event cameras, instead of conventional cameras, in structured-light (SL) systems improves the performance in 3D reconstruction and depth estimation due to their high temporal resolution, low latency, and high dynamic range. They enable the processing of data scanned at higher frequencies, a feature attributable to their high temporal resolution. Despite these advantages, the behaviour of the depth estimation algorithms at high scanning speeds in event-based SL systems is not well-studied in the existing literature. To address this gap, we present a study on the performance of a depth estimation algorithm using a custom hardware setup combining an event camera and a laser line galvo scanner. The setup provides the flexibility to record scenes at scanning frequencies up to 350 Hz. The proposed software pipeline utilizes single-plane interpolation for depth estimation and incorporates several pre- and post-processing techniques to reduce noise. We investigate the influence of varying the number of scans, scanning frequencies, and scene distance, on the software pipeline. From the experimental results, we conclude that the accuracy of the depth estimation algorithm decreases with increasing scanning speed, primarily due to noise. This study provides insights into the behaviour of the proposed SL system offering a platform to develop better algorithms for depth estimation in high-speed scanning scenarios. The dataset and source code for the software pipeline is also provided to facilitate further research in this field by using this study as a baseline.

This project was done in collaboration with SONY Research, Zurich, at Robotics Perception Group, University of Zurich.