Counter-UAS Integration

# ═══ DETECT DRONE RF WITH RTL-SDR ($25) + GNU RADIO ═══ # Install RTL-SDR driver + GNU Radio sudo apt install rtl-sdr gnuradio gr-osmosdr # Scan 2.4 GHz band (most drones) — needs an upconverter or # a HackRF One ($300) since RTL-SDR maxes at 1.7 GHz # With HackRF One — scan 2.4 GHz for drone video/control: hackrf_sweep -f 2400:2500 -w 500000 -l 32 -g 30 | \ python3 detect_peaks.py # detect_peaks.py: look for strong signals at known drone freqs # DJI: 2.400-2.4835 GHz (OcuSync/Lightbridge) # FPV: 5.650-5.925 GHz (analog/digital video) # ExpLRS: 2.400 GHz or 868/915 MHz # Crossfire: 868 MHz (EU) / 915 MHz (US)SDR Scan

# ═══ REMOTE ID RECEIVER (free detection layer) ═══ # Every compliant drone broadcasts ID on Bluetooth 5 + WiFi NaN # Use any phone with OpenDroneID app, or build a receiver: # ESP32-based Remote ID receiver (Arduino) # https://github.com/sxjack/uav_electronic_ids # or the official ASTM F3411 compatible: # https://github.com/opendroneid/receiver-android # Outputs: operator ID, drone serial, lat/lon/alt, speed, heading # Range: ~300m Bluetooth, ~1km WiFi NaN # Pipe to TAK via CoT for map display: # Set type="a-h-A" (hostile air) or "a-u-A" (unknown air)Remote ID

# ═══ ECHODYNE ECHOGUARD API — TRACK DRONE ON MAP ═══ # EchoGuard exposes tracks via REST API import requests, json def poll_radar(radar_ip="192.168.1.100"): resp = requests.get(f"http://{radar_ip}:8080/api/v1/tracks") tracks = resp.json() for track in tracks: lat = track['latitude'] lon = track['longitude'] alt = track['altitude_msl'] rcs = track['rcs_dbsm'] # Radar cross section (small = drone) vel = track['velocity_mps'] track_id = track['id'] # Filter: RCS < -10 dBsm = likely drone (not bird/aircraft) if rcs < -10: # Send to TAK as hostile air track: send_cot(lat, lon, alt, uid=f"radar-{track_id}", type="a-u-A") # unknown air # Poll at 1 Hz — radar updates faster but network is the bottleneck import time while True: poll_radar() time.sleep(1)Radar API

# ═══ YOLO DRONE DETECTION — CUSTOM TRAINED ═══ # Datasets for training drone detection models: # - Drone-vs-Bird: https://github.com/wosdetc/drone-vs-bird # - Anti-UAV: https://anti-uav.github.io/ # - DUT Anti-UAV: thermal + visible drone detection dataset # Train on your desktop GPU: yolo train model=yolov8s.pt data=drone_detection.yaml epochs=200 imgsz=640 # Classes to train: drone, bird, aircraft, helicopter, noise # The model learns to distinguish drones from birds at distance # Deploy on a PTZ camera with Jetson: # 1. Camera → GStreamer → YOLO inference # 2. Detection → PTZ auto-track (keep drone centered) # 3. Track data → TAK (hostile air icon on map) # For thermal: train separate model on LWIR drone signatures # Drones are hot (motors/batteries) against cold sky — easy detectAI Detection

# ═══ MULTI-SENSOR FUSION VIA ROS ═══ # Each sensor publishes to its own ROS topic: # /cuas/rf_detections — RF sensor (lat/lon/freq/signal strength) # /cuas/radar_tracks — Radar (lat/lon/alt/velocity/RCS) # /cuas/camera_detections — EO/IR AI (bounding box, class, confidence) # /cuas/remote_id — Remote ID (operator ID, serial, position) # /cuas/acoustic_bearing — Mic array (azimuth, elevation, confidence) # Fusion node subscribes to all, correlates by: # 1. Spatial proximity (detections within 50m = same target) # 2. Temporal correlation (detections within 2s window) # 3. Confidence weighting (radar > RF > camera > acoustic) # Output: /cuas/fused_tracks — high-confidence tracks with threat level # Publish to TAK for operator decision # Architecture: # ┌──────────┐ ┌──────────┐ ┌──────────┐ # │ RF Sensor│ │ Radar │ │ Camera+AI│ # └────┬─────┘ └────┬─────┘ └────┬─────┘ # ┌────▼──────────────▼──────────────▼─────┐ # │ ROS FUSION NODE │ # └────────────────┬────────────────────────┘ # ┌─────▼─────┐ # │ TAK/ATAK │ # └───────────┘Fusion

# ═══ INTERCEPTOR DRONE ARCHITECTURE ═══ # Hardware: Fast FPV-class quad (7"+ for speed) + companion computer # Sensors: Forward camera + AI (YOLO drone detection model) # Payload: Net launcher, or direct kinetic intercept (Anduril Anvil style) # Software stack on companion computer: # 1. Camera → YOLO drone detection (trained on drone silhouettes) # 2. Detection → visual servoing (keep target centered in frame) # 3. Visual servoing → velocity commands to FC via MAVROS # 4. Approach + capture/defeat at close range # ROS architecture: # /camera/image_raw → /yolo_detector → /target/bbox # /target/bbox → /visual_servo → /mavros/setpoint_velocity/cmd_vel # Key challenges: # - Target drone is also moving (pursuit dynamics) # - Need high FPS detection (30+ Hz) for tracking # - Prop wash turbulence at close range # - Legal authorization for kinetic intercept # See: Anduril Anvil (production system) # See: Fortem DroneHunter (net capture drone)Interceptor