Learning Monocular Depth from Events via Egomotion Compensation

TL;DR

This paper introduces an interpretable monocular depth estimation framework using event cameras, leveraging physical motion principles and novel modules to improve accuracy and robustness over existing black-box methods.

Contribution

It proposes a physics-inspired depth estimation approach with new modules for focus discrimination and cost aggregation, enhancing interpretability and performance.

Findings

Outperforms state-of-the-art methods by up to 10% in accuracy.

Effectively utilizes physical motion principles for depth estimation.

Demonstrates robustness in real-world and synthetic datasets.

Abstract

Event cameras are neuromorphically inspired sensors that sparsely and asynchronously report brightness changes. Their unique characteristics of high temporal resolution, high dynamic range, and low power consumption make them well-suited for addressing challenges in monocular depth estimation (e.g., high-speed or low-lighting conditions). However, current existing methods primarily treat event streams as black-box learning systems without incorporating prior physical principles, thus becoming over-parameterized and failing to fully exploit the rich temporal information inherent in event camera data. To address this limitation, we incorporate physical motion principles to propose an interpretable monocular depth estimation framework, where the likelihood of various depth hypotheses is explicitly determined by the effect of motion compensation. To achieve this, we propose a Focus Cost Discrimination (FCD) module that measures the clarity of edges as an essential indicator of focus level and integrates spatial surroundings to facilitate cost estimation. Furthermore, we analyze the noise patterns within our framework and improve it with the newly introduced Inter-Hypotheses Cost Aggregation (IHCA) module, where the cost volume is refined through cost trend prediction and multi-scale cost consistency constraints. Extensive experiments on real-world and synthetic datasets demonstrate that our proposed framework outperforms cutting-edge methods by up to 10\% in terms of the absolute relative error metric, revealing superior performance in predicting accuracy.