3D Reconstruction from Multiple Viewpoints Syllabus

Overview: This course provides a comprehensive introduction to 3D scene reconstruction using multiple camera views, combining theoretical foundations with hands-on implementation. Over 9 weeks, learners will explore multi-view geometry, depth estimation, and 3D modeling techniques through lectures, coding exercises, and real-world datasets. Expect to spend 6–8 hours per week on lectures, readings, and programming assignments.

Module 1: Camera Models and Calibration

Estimated time: 12 hours

• Pinhole camera model
• Intrinsic and extrinsic parameters
• Camera calibration techniques

Module 2: Stereo Vision and Depth Estimation

Estimated time: 12 hours

• Epipolar geometry and fundamental matrix
• Stereo correspondence algorithms
• Computing disparity and depth maps

Module 3: Structure from Motion

Estimated time: 18 hours

• Feature detection and matching across views
• Triangulation and 3D point estimation
• Bundle adjustment for refinement

Module 4: Practical Applications and Evaluation

Estimated time: 12 hours

• Reconstructing real-world scenes
• Assessing reconstruction quality
• Applications in robotics, AR/VR, and mapping

Module 5: Final Project

Estimated time: 20 hours

• Implement a complete 3D reconstruction pipeline
• Evaluate accuracy using real datasets
• Submit a technical report and reconstructed model

Prerequisites

• Familiarity with linear algebra and coordinate transformations
• Basic knowledge of Python and image processing
• Introductory understanding of computer vision concepts

What You'll Be Able to Do After

• Understand the fundamentals of multi-view geometry and camera calibration
• Compute depth maps from stereo image pairs
• Reconstruct 3D scenes using structure from motion techniques
• Apply triangulation and epipolar geometry to real-world datasets
• Evaluate the accuracy and limitations of 3D reconstruction pipelines