Learned Optimal Visual Time-of-Flight Imaging with Fisher Information Guidance.

Area of Science:

• Computer Vision
• Computational Imaging
• Robotics

Background:

• Indirect time-of-flight (iToF) imaging is vital for 3D scene understanding in robotics and augmented reality.
• Existing iToF systems struggle with low signal-to-noise ratio (SNR), especially in bright light, leading to depth inaccuracies and poor edge reconstruction.

Purpose of the Study:

• To develop an optimized iToF imaging scheme for improved depth accuracy and robustness under challenging low-SNR conditions.
• To enhance geometric detail preservation and edge reconstruction in 3D depth maps.

Main Methods:

• An end-to-end learning approach was used to optimize iToF coding functions and depth reconstruction, guided by Fisher information.
• RGB image data was integrated to improve the robustness of iToF coding functions against noise.
• A dual-branch network was designed to leverage visual edge information for precise depth estimation and detail preservation.

Main Results:

• The proposed method significantly improves depth accuracy and robustness in low-SNR environments.
• Fine geometric details and depth discontinuities are better preserved compared to traditional methods.
• Experiments on synthetic and real-world data validate the effectiveness of the visual iToF imaging scheme.

Conclusions:

• The integrated visual iToF imaging scheme offers a robust solution for accurate 3D depth sensing in challenging illumination.
• This approach enhances the reliability of iToF technology for critical applications like autonomous driving and robotics.