Real-Time Object Detection for Autonomous Drones

Project Overview  Develop a fully autonomous drone capable of detecting, classifying, and avoiding moving/static obstacles (people, vehicles, birds, trees, power lines) in real-time using only onboard vision — no GPS or remote pilot required. The system runs entirely on edge hardware (NVIDIA Jetson Orin Nano / NX) and achieves >45 FPS detection with >85% accuracy even on tiny objects from 50–150 meters altitude. Key Features Real-time object detection using YOLOv11n/s (Ultralytics 2025) Multi-class detection: person, car, bicycle, dog, bird, powerline, building Depth-aware obstacle avoidance using stereo camera or LiDAR fusion Dynamic path replanning with A* or DWA (Dynamic Window Approach) Emergency hover/land on critical obstacle detection Works in GPS-denied indoor/outdoor environments Optional thermal camera support for night/search-and-rescue missions Hardware Requirements (Budget-Friendly Options) Drone frame: F450/F550 or S500 Flight controller: Pixhawk 6X or Cube Orange (with ArduPilot/PX4) Companion computer: NVIDIA Jetson Orin Nano 8GB (~$500) or Jetson Nano 4GB (~$99 used) Camera: Intel RealSense D435i (stereo + depth) or Raspberry Pi HQ Camera + ZED 2i Optional: FLIR Lepton thermal module Total cost: $800–$1500 Software Stack (All Open-Source) ROS2 Humble/Foxy (Robot Operating System) Ultralytics YOLOv11 (pre-trained + custom trained) OpenCV + TensorRT for optimized inference MAVROS/MAVLink for drone control Gazebo + PX4 SITL for simulation testing Why This Project Stands Out in 2025 Solves real-world problems: urban delivery, indoor warehouse navigation, disaster response Uses latest YOLOv11 (released Oct 2024) — better than YOLOv8/v10 on small aerial objects 100% onboard processing — no cloud dependency High publication potential (IEEE Transactions, ICRA, IROS) Winning project material for competitions (SAE Aero Design, DroneTech, Hackathons) Expected Results Detection accuracy: >87% mAP@0.5 on VisDrone dataset Inference speed: 48–65 FPS on Jetson Orin Nano Obstacle avoidance success rate: >95% in simulated and real tests Safe flight in cluttered environments (trees, people, buildings) Applications Amazon-style package delivery drones Search & rescue in disaster zones Power line inspection with automatic obstacle avoidance Indoor inventory drones in warehouses Military/Police surveillance in urban areas Bonus Advanced Versions (For M.Tech/PhD Level) Add multi-drone swarm coordination Implement active tracking of moving humans/vehicles Thermal + RGB fusion for night operations Privacy-preserving detection (blur faces automatically) 5G-enabled offloading for extreme compute tasks