Skip to content

System_Overview.md

Latest commit

 

History

History
51 lines (38 loc) · 2.52 KB

System_Overview.md

File metadata and controls

51 lines (38 loc) · 2.52 KB

System Architecture and Overview

Proposed System

The proposed system presents a fully simulated autonomous drone framework that integrates LiDAR-based odometry, optical flow-derived motion cues, and deep learning perception for environment-aware navigation. The complete pipeline is executed in ROS 2 and Gazebo, facilitating controlled testing without physical hardware.

Core Architecture Components

# Component Technology
1 Sensor Simulation Gazebo (LiDAR + RGB Camera)
2 Optical Flow Computation Visual motion vectors between frames
3 Multi-Sensor Fusion Factor Graph Optimization (FGO)
4 Object Detection YOLOv8 (Darknet ROS integration)
5 Scenario Validation Disaster-mimicking Gazebo worlds

Software Stack

Category Tools & Frameworks
OS Ubuntu 18.04 / 20.04 / 22.04 LTS · WSL2
Middleware ROS 2 Humble
Simulation Gazebo Garden · ArduPilot SITL
Perception YOLOv8 · OpenCV · PyTorch
Localization LiDAR-SLAM · Google Cartographer
Languages Python 3 · C++17
Standards IEEE 802.15.4 · ISO 12100 · ROS REP

Working Methodology

Sensor Simulation Setup

The Gazebo environment hosts a virtual UAV equipped with LiDAR and an RGB camera. All sensor data are published via ROS topics (/scan, /camera/image_raw), with strict temporal synchronization.

Optical Flow Estimation

Motion information is derived from consecutive image frames using visual motion constraints, yielding an instantaneous velocity field representing drone motion.

Sensor Fusion via Factor Graph Optimization (LOFF)

The system mimics LOFF behavior by fusing LiDAR odometry (point cloud alignment) with optical motion estimation within a Factor Graph Optimization framework. This minimizes overall error across accumulated constraints, yielding a robust, consistent fused pose estimate.

Object Detection via YOLOv8

YOLOv8 processes camera frames to detect and classify objects (humans, vehicles). Bounding box outputs are integrated with the localization system to implement perception-driven feedback — adjusting trajectory to avoid collisions and triggering intelligent landing behavior.

Compliance Standards

  • Communication & Data Transfer: IEEE 802.15.4, IEEE 1722, ROS REP Standards
  • Safety & Compliance: ISO 12100, ASTM F3201, IEEE 829
  • Technical Precision: IEEE 754 (floating-point precision)
  • Simulation Standards: Gazebo SDF & ArduPilot SITL modeling