Tackling 3D ToF Artifacts Through Learning and the FLAT Dataset

TL;DR

This paper presents a deep-learning method to reduce artifacts in 3D ToF camera depth data, supported by a new synthetic dataset, FLAT, enabling improved reconstruction accuracy on real and simulated data.

Contribution

It introduces a novel two-stage deep-learning approach and the FLAT dataset for comprehensive artifact correction in ToF depth sensing.

Findings

Improved depth reconstruction accuracy over state-of-the-art methods

Effective artifact removal on both simulated and real data

FLAT dataset captures diverse nonidealities in ToF measurements

Abstract

Scene motion, multiple reflections, and sensor noise introduce artifacts in the depth reconstruction performed by time-of-flight cameras. We propose a two-stage, deep-learning approach to address all of these sources of artifacts simultaneously. We also introduce FLAT, a synthetic dataset of 2000 ToF measurements that capture all of these nonidealities, and allows to simulate different camera hardware. Using the Kinect 2 camera as a baseline, we show improved reconstruction errors over state-of-the-art methods, on both simulated and real data.