Autonomous Airborne Security System Development for urgent
Budget: $30 – $250 USD
Project Overview
We are building an autonomous airborne security system that patrols a fenced compound, detects fence breaches or suspicious humans in real time, and notifies QGroundControl (QGC).
The system will run fully in simulation first, then transition to hardware-in-the-loop (HITL). Core tech:
• Simulation: PX4 autopilot (latest stable tag) + Gazebo Harmonic (a.k.a. “new Gazebo”) + QGC.
• Environment: High-fidelity urban scene authored in Blender (provided as .dae / .blend)—must be imported into Gazebo with terrain, buildings, gates, trees, light sources.
• Perception: YOLOv8 (Ultralytics) for real-time detection of people and fence-breach events, running on Python NVIDIA Jetson Nano profile (GPU acceleration enabled).
• UAV Platform: Aurelia X6 Standard hexacopter model, configured in PX4 with realistic mass/inertia, battery, and performance envelope.
• Sensors: Two on-board cameras—Arducam IMX519 (RGB, auto-focus) and FLIR Lepton 3.5 (thermal). Data must stream into your CV pipeline.
• Mission Logic:
• pre-programmed perimeter patrol via waypoints;
• stop-and-hover on detected fence breach;
• person-follow mode with yaw tracking if a human is detected inside the perimeter;
• MAVLink custom messages to QGC (visual alert, geofence breach, human ID).
• Programming language: 100 % Python (ROS-free—we talk directly to Gazebo Transport / MAVSDK).
⸻
Work Packages & Deliverables
1. Environment & Toolchain Setup
• Install PX4, Gazebo Harmonic, QGroundControl on Ubuntu 22.04.
• Validate PX4-Gazebo SITL link (make px4_sitl gazebo), confirm MAVLink streaming to QGC.
• Document prerequisites (gcc, ninja, path tweaks, udev rules).
2. Scene Import & Asset Preparation
• Convert the supplied Blender scene into Gazebo (USD/SDF + textures).
• Bake collision meshes, LODs, lighting.
• Provide a run-ready world file compound.world.
3. UAV & Sensor Modeling
• Configure aircraft parameters according to Aurelia X6 spec sheet (mass, inertia, rotor positions, thrust curve, battery 6S 20 Ah).
• Add two camera sensors: forward RGB (1080p 30 fps) and downward FLIR (160×120 8.7 Hz) with realistic latency/noise.
4. Computer Vision Integration
• Containerized YOLOv8 (PyTorch) pipeline with GStreamer sink pulling frames from Gazebo topics.
• Real-time inference ≤ 50 ms per frame on desktop GPU (Jetson profile documented).
• JSON/MAVLink output: FENCE_BREACH, PERSON_DETECTED, bounding box coordinates.
5. Mission & Autonomy Scripts
• Autonomous patrol (imported from QGC plan or off-board Python script).
• State machine: Patrol → Breach Detected → Hover/Follow → Resume.
• Custom MAVLink messages displayed in QGC UI with severity & audio cue.
6. Dynamic Entities
• Spawn walking human actors along parametric paths.
• Spawn/remove fence-panel links to emulate “hole” events.
• Trigger CV events and verify system response.
7. Testing & Metrics
• 10-minute demo flight showing at least two breach events and one human follow.
• Provide log analysis (ULog + vision timestamps) proving <2 s end-to-end alert latency.
8. Documentation & Handoff
• Step-by-step README for clean install on fresh Ubuntu.
• Video walkthrough (~5 min) of starting the whole stack, viewing alerts in QGC, and stopping.
• Optional: short guide for moving to HITL with Pixhawk 6X.
We are building an autonomous airborne security system that patrols a fenced compound, detects fence breaches or suspicious humans in real time, and notifies QGroundControl (QGC).
The system will run fully in simulation first, then transition to hardware-in-the-loop (HITL). Core tech:
• Simulation: PX4 autopilot (latest stable tag) + Gazebo Harmonic (a.k.a. “new Gazebo”) + QGC.
• Environment: High-fidelity urban scene authored in Blender (provided as .dae / .blend)—must be imported into Gazebo with terrain, buildings, gates, trees, light sources.
• Perception: YOLOv8 (Ultralytics) for real-time detection of people and fence-breach events, running on Python NVIDIA Jetson Nano profile (GPU acceleration enabled).
• UAV Platform: Aurelia X6 Standard hexacopter model, configured in PX4 with realistic mass/inertia, battery, and performance envelope.
• Sensors: Two on-board cameras—Arducam IMX519 (RGB, auto-focus) and FLIR Lepton 3.5 (thermal). Data must stream into your CV pipeline.
• Mission Logic:
• pre-programmed perimeter patrol via waypoints;
• stop-and-hover on detected fence breach;
• person-follow mode with yaw tracking if a human is detected inside the perimeter;
• MAVLink custom messages to QGC (visual alert, geofence breach, human ID).
• Programming language: 100 % Python (ROS-free—we talk directly to Gazebo Transport / MAVSDK).
⸻
Work Packages & Deliverables
1. Environment & Toolchain Setup
• Install PX4, Gazebo Harmonic, QGroundControl on Ubuntu 22.04.
• Validate PX4-Gazebo SITL link (make px4_sitl gazebo), confirm MAVLink streaming to QGC.
• Document prerequisites (gcc, ninja, path tweaks, udev rules).
2. Scene Import & Asset Preparation
• Convert the supplied Blender scene into Gazebo (USD/SDF + textures).
• Bake collision meshes, LODs, lighting.
• Provide a run-ready world file compound.world.
3. UAV & Sensor Modeling
• Configure aircraft parameters according to Aurelia X6 spec sheet (mass, inertia, rotor positions, thrust curve, battery 6S 20 Ah).
• Add two camera sensors: forward RGB (1080p 30 fps) and downward FLIR (160×120 8.7 Hz) with realistic latency/noise.
4. Computer Vision Integration
• Containerized YOLOv8 (PyTorch) pipeline with GStreamer sink pulling frames from Gazebo topics.
• Real-time inference ≤ 50 ms per frame on desktop GPU (Jetson profile documented).
• JSON/MAVLink output: FENCE_BREACH, PERSON_DETECTED, bounding box coordinates.
5. Mission & Autonomy Scripts
• Autonomous patrol (imported from QGC plan or off-board Python script).
• State machine: Patrol → Breach Detected → Hover/Follow → Resume.
• Custom MAVLink messages displayed in QGC UI with severity & audio cue.
6. Dynamic Entities
• Spawn walking human actors along parametric paths.
• Spawn/remove fence-panel links to emulate “hole” events.
• Trigger CV events and verify system response.
7. Testing & Metrics
• 10-minute demo flight showing at least two breach events and one human follow.
• Provide log analysis (ULog + vision timestamps) proving <2 s end-to-end alert latency.
8. Documentation & Handoff
• Step-by-step README for clean install on fresh Ubuntu.
• Video walkthrough (~5 min) of starting the whole stack, viewing alerts in QGC, and stopping.
• Optional: short guide for moving to HITL with Pixhawk 6X.