Real-time 3D reconstruction from single-photon lidar data using plug-and-play point cloud denoisers

Area of Science:

• Optics and Photonics
• Computer Vision
• Computational Imaging

Background:

• Single-photon lidar offers potential for depth imaging in challenging environments.
• A key limitation has been the extensive time required for data analysis.
• Real-time processing is crucial for practical lidar applications.

Purpose of the Study:

• To introduce a novel computational framework for real-time 3D scene reconstruction from single-photon lidar data.
• To overcome the time limitations associated with analyzing single-photon lidar data.
• To enable robust 3D imaging through cluttered environments and moving scenes.

Main Methods:

• Developed a computational framework integrating statistical models and scalable computer graphics tools.
• Applied the framework to process single-photon lidar data acquired in daylight up to 320 meters.
• The method handles an unknown number of surfaces per pixel for enhanced target detection.

Main Results:

• Achieved real-time 3D scene reconstruction with processing times around 20 milliseconds.
• Demonstrated successful reconstruction of complex outdoor scenes.
• Enabled imaging through cluttered scenes by handling multi-surface pixels.

Conclusions:

• The new framework enables real-time 3D reconstruction from single-photon lidar data.
• This advancement significantly reduces processing time, making lidar practical for complex applications.
• Paves the way for single-photon lidar operating at video rates for real-world 3D imaging.