AR SLAM System

A real-time monocular visual SLAM (Simultaneous Localization and Mapping) system built from scratch in C++20, demonstrating core augmented reality tracking capabilities similar to those used in ARCore and ARKit.

Overview

This project implements fundamental AR tracking technology without using existing SLAM libraries, built specifically to demonstrate understanding of computer vision and AR. The system achieves real-time performance (46+ FPS) while tracking 1000+ features simultaneously with 3D visualization.

Key Features

• Real-time Performance: 1000+ tracked features
• 3D Point Cloud Visualization: OpenGL-based rendering with depth-based coloring
• Robust Feature Tracking: ORB feature detection with KLT optical flow
• Modern C++ Design: C++20 with smart pointers, templates, and RAII principles
• Cross-platform Ready: Designed for Linux with V4L2 camera interface structure
• Real-time operation on standard hardware
• Robust tracking with automatic recovery
• Handles typical handheld camera motion
• Automatic re-detection when tracking degrades
• Memory-efficient design suitable for embedded systems

System Architecture

ar-slam-system/ 
├── Core Components (src/core/) 
│   ├── Frame Management 
│   ├── Feature Detection & Tracking 
│   └── Memory Management 
├── 3D Visualization (src/rendering/) 
│   ├── OpenGL 3.3 Core Profile 
│   └── Real-time point cloud rendering 
└── Testing Framework (tests/) 
    ├── Performance benchmarks 
    └── Unit tests 

Quick Start

Prerequisites

sudo apt-get update 
sudo apt-get install -y build-essential cmake libopencv-dev libglew-dev libglfw3-dev libglm-dev libeigen3-dev 

Build Instructions

# Clone repository 
git clone https://github.com/Maxencejules/ar-slam-system.git 
cd ar-slam-system 

# Create build directory 
mkdir build && cd build 

# Configure with CMake (Release mode for best performance) 
cmake .. -DCMAKE_BUILD_TYPE=Release 

# Build with all cores 
make -j$(nproc) 

# Run the main 3D tracking demo 
./src/camera_3d 

Usage

Controls

Key	Action
Arrow Keys	Rotate/zoom 3D view
Space	Print performance statistics
ESC/Q	Exit application

Output Windows

1. 2D Camera View: Live camera feed with tracked features (green dots), motion trails, and performance stats.
2. 3D Point Cloud View: Spatial visualization with depth-based coloring and reference grid floor.

GitHub Repository

Technical Implementation

Feature Detection & Tracking

• ORB Feature Detection: Up to 1000 features per frame
• KLT Optical Flow: Pyramidal Lucas-Kanade with 3 levels
• RANSAC: Outlier rejection with fundamental matrix estimation (Epipolar Geometry Constraint)

Memory Pool Design

Custom pool allocator with O(1) allocation/deallocation, maintaining 256MB constraint for embedded systems.

3D Visualization

OpenGL 3.3 Core Profile with custom shaders for point cloud rendering.

• Depth Coloring: Near (Red) -> Mid (Green) -> Far (Blue)
• Spatial Context: 3D grid floor for ground plane reference

Dependencies

Library	Version	Purpose
OpenCV	4.6.0	Computer vision algorithms
OpenGL	3.3+	3D rendering
GLFW	3.3+	Window management
GLEW	2.2+	OpenGL extensions
GLM	0.9.9+	Mathematics
Eigen3	3.4+	Linear algebra

Project Structure

ar-slam-system/ 
├── CMakeLists.txt 
├── README.md 
├── include/ 
│   ├── core/ 
│   │   ├── frame.h 
│   │   ├── feature_tracker.h 
│   │   └── memory_pool.h 
│   └── rendering/ 
│       └── gl_viewer.h 
├── src/ 
│   ├── camera_3d_test.cpp 
│   ├── core/ 
│   │   ├── frame.cpp 
│   │   └── feature_tracker.cpp 
│   └── rendering/ 
│       └── gl_viewer.cpp 
└── tests/ 
    ├── benchmark/ 
    │   └── benchmark_main.cpp 
    └── unit/ 
        └── test_features.cpp 

Future Enhancements

• Bundle adjustment for global optimization
• Loop closure detection
• Dense reconstruction
• IMU sensor fusion
• Mobile deployment (Android NDK)

License

MIT License - see LICENSE file for details.

Contact

Maxence Jules

• Email: powe840@gmail.com
• LinkedIn: linkedin.com/in/julesmax
• GitHub: @Maxencejules

Built with passion for augmented reality and computer vision.