Vision-Based 3D Object Tracking with Hough Transform in ROS

Overview

This master’s thesis presents a vision-based system for real-time 3D object coordinate estimation and tracking. The approach leverages the Hough Transform for robust object detection combined with a modular 3D estimation pipeline implemented entirely in ROS (Robot Operating System).

Key Features

• Monocular 3D Estimation: Extracts 3D object coordinates from a single camera input without requiring depth sensors
• Hough Transform Detection: Uses the Generalized Hough Transform for robust shape-based object detection
• Modular ROS Architecture: Cleanly separated ROS nodes for detection, estimation, and tracking
• Real-Time Performance: Optimized pipeline capable of running at interactive frame rates
• Robotic Integration: Designed for direct integration with robotic manipulation and navigation stacks

System Architecture

The system consists of three main modules:

1. Detection Module: Processes camera frames using Hough Transform to identify and localize objects in 2D
2. 3D Estimation Module: Converts 2D detections to 3D world coordinates using camera calibration and geometric reasoning
3. Tracking Module: Maintains object identity and trajectory across frames using state estimation

Technologies Used

• ROS (Robot Operating System)
• OpenCV for image processing
• Python / C++
• Hough Transform algorithms
• Camera calibration tools
• Point cloud processing

Publication

This work was published as a preprint on TechRxiv:

Vision-Based 3D Object Coordinate Estimation and Tracking Using Hough Transform and Modular 3D Estimation in ROS

Contributors

• Oussama Errouji
• Imad-Eddine NACIRI
• Jade Bousliman