Obstacle Detection and Avoidance System for Unmanned Aerial Vehicles Based on Monocular Camera

Abstract

Unmanned Aerial Systems (UAS), commonly known as drones, are aircraft systems without a human pilot onboard, controlled remotely or autonomously. Algorithms like YOLO (You Only Look Once) for object detection and pathfinding algorithms like A* (A-Star) can quickly navigate around large, static objects like buildings or trees. However, detecting small objects and handling dynamic aerial environments remain challenging. To address this, we introduce an innovative system for small object detection and real-time path planning using a monocular camera. Our dual-stage system combines traditional detection methods like background subtraction with advanced deep-learning techniques for improved reliability to create initial detection zones, further refined by target tracking methods for increased accuracy and depth predictor for getting estimated distance. Additionally, we have developed a new path planning algorithm, Circle Rapidly-exploring Random Trees-star (Circle RRT*), for effective obstacle avoidance. Our Obstacle Detection and Avoidance architecture navigates dynamic conditions with greater precision and speed in identifying small targets.