Computer Vision System for Object Detection, Classification, Localization, and Partial 3D Reconstruction
This repository presents the structure of my undergraduate thesis in Electronics Engineering at Universidad Simón Bolívar, focused on Computer Vision, Depth Estimation, and 3D Reconstruction.
The thesis received an Outstanding Mention from the IEEE Ambassador for Venezuela, who described it as “at a postgraduate level.”
The full document is currently under academic review; therefore, only the index is publicly shared here.
- Motivation
- Objectives
- General Objective
- Specific Objectives
- Structure
- Definition of a Digital Image
- Image Formation: The Pinhole Principle
- Pinhole Camera Model and Perspective Projection
- Coordinate Systems: Image, Camera, and World
- Camera Lens and Types of Distortion
- Homogeneous Coordinates and Linear Camera Model
- Types of Calibration
- Direct Linear Transformation (DLT) Method
- Homography
- Zhang Method
- Technical Specifications of the Camera Used
- Chessboard Calibration
- Calibration Results with Chessboard
- ArUco Markers and ChArUco Calibration
- Definition
- Marker Creation and Detection
- ChArUco Boards
- Calibration Results with ChArUco
- Pattern Comparison
- Calibration Evaluation Experiment
- Neural Networks
- Linear Regression
- The Perceptron
- The Multilayer Perceptron (MLP)
- Convolutional Neural Networks (CNNs)
- Classical Filters in Image Processing
- Basic CNN Architecture
- Comparison Between Classical and Learned Filters
- You Only Look Once (YOLO) Model
- General Description
- Training Phase
- Inference Phase
- Training with Roboflow Dataset and YOLOv10 for Object Detection
- Active Techniques
- Structured Light
- Photometric Stereo
- Time of Flight (ToF)
- LED Line Scanner (Laser Scanner)
- Passive Techniques
- Neural Networks
- Depth from Defocus
- Shape from Shadows
- Shape from Specularities
- Shape from Texture
- Shape from Silhouettes
- Stereo Vision
- Structure from Motion
- Technique Selection Process for the Project
- Emulated Stereo Vision
Chapter 6: Own Implementation – Conversion of Relative to Absolute Depth Maps with Reference Markers (CRA2M)
- Overview of MiDaS Functioning
- Datasets Used for Training and Evaluation
- MiDaS Limitations
- Introduction to Transformer Technology
- Using Transformers to Improve MiDaS
- Version Comparisons
- CRA2M Explanation
- Obtaining the Relative Depth Map
- ArUco Marker Detection and Translation Vector Extraction
- Extracting the Marker’s Relative Depth Value from the Depth Map
- Scale Factor Calculation
- Applying the Scale Factor to the Depth Map
- Results
- Homogeneous Background
- Non-homogeneous Background
- Multiple Iterations for Averaging Results
- Using Chessboard Pattern for Reference Distance
- Processing Time Difference Between ArUco and Chessboard Methods
- Lighting Impact on ArUco Detection
- Conclusions and Optimization Recommendations
- Evolution Toward Gray-Code Structured Light
- Depth Estimation Using a Laser Pointer
- Depth Estimation Using a Light Stripe
- Depth Estimation with Multiple Light Stripes
- Binary-coded Structured Light
- Structured Light with Gray Coding
- Pixel Classification and Direct/Indirect Light Separation
- Simplified Illumination Model
- Steps for Illumination Decomposition
- Difference Between Armstrong and Nayar Methods
- Formula for Determining the Number of Patterns to Use
- Xu and Aliaga Pixel Classification Rules
- Stereo Calibration
- Epipolar Geometry
- Fundamental Matrix Calculation
- Projector-Camera System Calibration
- Triangulation
- Technical Aspects of Projectors to Consider
- Projection Technologies
- Projector Technical Specifications
- Results
- Pixel Encoding
- Pixel Decoding
This chapter presents an innovative 3D reconstruction method currently under submission to the International Conference on Automation, Robotics and Applications (ICARA 2026), Istanbul, Turkey.
- General Conclusions
- Future Research Directions
- Python, OpenCV, PyTorch
- YOLOv10, MiDaS, Transformers
- Camera Calibration (Zhang Method, DLT, Homographies)
- Depth Estimation, Stereo Vision, Structure from Motion
- 3D Reconstruction and Structured Light Systems
This index is shared for academic and professional reference.
The detailed content of Chapter 8 remains confidential until the ongoing publication process is complete.