Obstacle Detection and Avoidance System Based on Monocular Camera and Size Expansion Algorithm for UAVs.

Area of Science:

• Robotics
• Computer Vision
• Autonomous Systems

Background:

• Designing robust real-time obstacle detection and avoidance systems is critical for autonomous aerial vehicles (AAVs).
• Micro and small AAVs face Size, Weight, and Power (SWaP) constraints, making lightweight sensors like digital cameras preferable over laser or radar.
• Existing methods often rely on stereo cameras for 3D obstacle mapping, which can be computationally intensive.

Purpose of the Study:

• To propose a novel, lightweight, real-time obstacle detection and avoidance system for AAVs using a monocular camera.
• To mimic human-like collision detection by analyzing visual cues of approaching obstacles.
• To enable safe navigation for AAVs under SWaP constraints.

Main Methods:

• The algorithm analyzes changes in the size of detected feature points on obstacles across consecutive frames.
• It calculates the expansion ratios of the convex hull formed by these feature points to assess proximity.
• Obstacles with a high probability of collision are identified by comparing their area ratio with the UAV's position, and avoidance maneuvers are executed based on estimated 2D positions and waypoints.

Main Results:

• The proposed monocular camera-based system accurately detects and predicts potential collisions with obstacles.
• Real-world indoor and outdoor flight tests demonstrated the system's effectiveness.
• The algorithm shows comparable or superior accuracy to existing related works.

Conclusions:

• The developed method offers an effective and lightweight solution for real-time obstacle detection and avoidance in AAVs.
• It provides a viable alternative to more complex stereo vision systems, especially for SWaP-constrained platforms.
• The system enhances the safety and robustness of autonomous aerial navigation.