PS-EIP: Robust Photometric Stereo Based on Event Interval Profile

TL;DR

PS-EIP advances event-based photometric stereo by constructing an Event Interval Profile that captures temporal relationships between adjacent event intervals. By fitting a theoretically derived EIP to observed profiles and applying shape-based masks to remove non Lambertian distortions, it achieves robust pixelwise normal recovery without deep learning. Quantitative results on 21 3D printed objects show substantial improvements over EventPS-FCN, with an average angular error around 8.12°, demonstrating resilience to shadows and specularities. The approach preserves detail, operates on CPU without GPUs, and offers practical gains for energy-efficient surface reconstruction in challenging lighting and material conditions.

Abstract

Recently, the energy-efficient photometric stereo method using an event camera has been proposed to recover surface normals from events triggered by changes in logarithmic Lambertian reflections under a moving directional light source. However, EventPS treats each event interval independently, making it sensitive to noise, shadows, and non-Lambertian reflections. This paper proposes Photometric Stereo based on Event Interval Profile (PS-EIP), a robust method that recovers pixelwise surface normals from a time-series profile of event intervals. By exploiting the continuity of the profile and introducing an outlier detection method based on profile shape, our approach enhances robustness against outliers from shadows and specular reflections. Experiments using real event data from 3D-printed objects demonstrate that PS-EIP significantly improves robustness to outliers compared to EventPS's deep-learning variant, EventPS-FCN, without relying on deep learning.

Figures (18)

• Figure 1: Procedure of the proposed method: Events are recorded under moving light conditions. The profile is reconstructed from the inverse of the event intervals. A surface normal is determined through curve fitting for each pixel. The proposed method is robust to non-Lambertian effects such as specularity and shadows.
• Figure 2: Reconstruction of a profile from a series of events of a single cycle light movement. (a) Recorded events. (b) Reconstructed event interval profile (EIP).
• Figure 3: Designs of the masks for non-Lambertian effects. The left: intensity-based image from events, $\Delta I$. The middle and right: reconstructed profiles at the red and orange circle points in $\Delta I$, respectively. The gray dashed line represents the corresponding ideal profile. The gray fill indicates the masked regions to be removed.
• Figure 4: Implementation. (a) Prototype system. (b) Top: Color checker. Middle and bottom: Event accumulation images under a light source modulated by a triangular wave. The power range of the light source is from $10$ to 20% for the middle and from $50$ to 100% for the bottom ($k=2$).
• Figure 5: Analysis on the number of measurements for the average. (a) Averaged profiles for the different numbers of measurements. (b) Mean angular errors w.r.t. the number of measurements when targeting the three sphere-based objects.