3D Scene Reconstruction

A MATLAB-based app for real-time 3D scene reconstruction from interior image sequences using geometric computer vision algorithms (group project for the EI70110 challenge, SS23 Edition).

Table of Contents

• Description
• Requirements
• Quick Start

Description

This software processes a collection of room images to generate a 3D model, reconstructing its spatial structure and providing an abstract visual representation. While it can work with random images, sequentially ordered images yield more accurate results.

Requirements

This software has been developed and tested using the following software and libraries:

• MATLAB (version R2023a)
• Computer Vision Toolbox
• Image Processing Toolbox
• Statistics and Machine Learning Toolbox

Quick Start

Step 1: Launch GUI

Ensure MATLAB R2023a is running. Execute the main.m script to open the GUI:

Step 2: Load Camera Parameters

Click "Load Parameters" and select the folder containing images.txt and cameras.txt.

Step 3: Load Images

Click "Load Images" to select the images of the room you want to process.

Step 4: Select Feature Extraction Algorithm

Choose between "SIFT" (more accurate but slower) and "SURF" for feature extraction.

Step 5: Generate Point Cloud

Click "Generate Point Cloud" to create and display a 3D point cloud of the room.

Step 6: Optimize Parameters

Adjust the following parameters for optimal results:

• Distance for Denoising: Controls clustering distance during denoising.
• Point Count for Denoising: Sets threshold for removing noise.
• Shrink Factor for Room Shape: Adjusts the fit of the wall structure.
• Number of Boxes in Visualization: Modifies point clusters created during k-means.

Step 7: Visualize the 3D Model

Click "Plot Visualization" to view the abstract 3D model. Click again to return to the point cloud view.

Step 8: Measure Distance

Enable "Distance Measurement Mode" to select two points in the 3D model and calculate the distance between them.