ROS-based ground stereo vision detection: implementation and experiments.

Area of Science:

• Robotics
• Computer Vision
• Autonomous Systems

Background:

• Autonomous landing of unmanned aerial vehicles (UAVs) requires robust object detection in complex environments.
• Ground stereo vision systems offer a viable solution for precise navigation and landing guidance.
• Cluttered scenes pose significant challenges for current flying object detection algorithms.

Purpose of the Study:

• To implement and validate an open-source flying object detection system for UAVs.
• To detail the architecture and workflow of a ground stereo vision guidance system.
• To evaluate the Chan-Vese detection algorithm within a Robot Operating System (ROS) environment.

Main Methods:

• The Chan-Vese detection algorithm was implemented in the ROS environment.
• A data-driven interactive scheme was developed for dataset collection and parameter tuning.
• Stereo sequential images and sensor data (pan-and-tilt unit, onboard sensors, differential GPS) were collected during outdoor flight experiments.

Main Results:

• The ROS-based detection algorithm demonstrated effectiveness in cluttered scenes.
• Experimental results validated the system's performance using a custom-collected dataset.
• The ground stereo vision guidance system provided reliable data for autonomous landing applications.

Conclusions:

• The developed open-source system provides an effective solution for flying object detection in cluttered environments.
• The implementation within ROS facilitates integration and further development for UAV applications.
• The study validates the use of stereo vision and the Chan-Vese algorithm for autonomous landing guidance.