Monocular visual SLAM, visual odometry, and structure from motion methods applied to 3D reconstruction: A...

Area of Science:

• Computer Vision
• Robotics
• 3D Reconstruction

Background:

• Monocular Simultaneous Localization and Mapping (SLAM), Visual Odometry (VO), and Structure from Motion (SFM) are key techniques for 3D reconstruction using monocular cameras.
• These visual-only methods are favored for their cost-effectiveness, portability, and widespread availability on handheld devices.

Purpose of the Study:

• To provide a comprehensive overview and taxonomy of monocular SLAM, VO, and SFM techniques for 3D reconstruction.
• To systematize existing methods, including classic and machine learning approaches, by detailing their algorithms and formulations.

Main Methods:

• An extended taxonomy was developed, encompassing classic, machine learning, direct, indirect, dense, and sparse methods.
• A detailed review of 42 methods (18 classic, 24 machine learning) was conducted, analyzing their algorithms and formulations.
• Algorithms were summarized based on specific criteria to aid in system selection and design.

Main Results:

• The study classified 42 monocular 3D reconstruction methods within an extended taxonomy.
• Key information for 18 classic and 24 machine learning methods was systematically presented.
• An analysis of temporal trends revealed the sustained popularity of classical-sparse-indirect and classical-dense-indirect approaches.

Conclusions:

• The research offers a structured understanding of monocular 3D reconstruction techniques.
• The findings provide valuable insights for researchers and developers in selecting or designing 3D reconstruction systems.
• Classical-sparse-indirect and classical-dense-indirect methods remain the most prevalent solutions in the field.