NIGHT OPERATIONS
FPV AND ISR IN DARKNESS

Thermal vs Visual at Night

Visual cameras are useless at night without illumination. Thermal cameras see heat — they work in total darkness, through smoke, and through light fog. The trade-off: thermal resolution is low (256×192 vs 4056×3040 visual), so fine identification (reading markings, distinguishing vehicle models) is difficult. Thermal excels at DETECTION (is something there?) while visual excels at IDENTIFICATION (what exactly is it?). At night, use thermal for detection and close-approach visual (with IR illumination) for identification.

FPV Night Flying Procedure

The transition from day to dusk operations requires a deliberate handover procedure. Pilots switch their primary display from visual to infrared, recalibrate their perception of speed and distance in the monochrome image, and verify that the AI model loaded on the Jetson matches the lighting conditions. The dedicated infrared model has different confidence thresholds and classification parameters optimized for the contrast ratios found in low-ambient-temperature environments typical of Scandinavian operations.

Thermal Detection Advantage

AI detection performance improves at night. During daylight, solar radiation heats every surface — rocks, roads, metal roofs, parked vehicles — creating cluttered thermal background. At night, all passively heated surfaces cool to ambient within 2-3 hours. Only objects with internal heat sources remain warm: running engines (80-120°C against 0-10°C background), generators, heated buildings. YOLOv8 thermal model achieves mAP 0.82 at night versus 0.75 during day because reduced clutter produces cleaner imagery with fewer false positives. The implication: Fischer 26 ISR is MORE effective during darkness, not less — the optimal ISR window is 2-4 hours after sunset when thermal contrast is maximum.

IR Illumination for Visual Navigation

Fischer 26 visual camera (IMX477) requires illumination to function at night. Visible light (white LED) would reveal the drone's position to ground observers. Solution: 850nm infrared LED illuminator (invisible to the naked eye) provides illumination for the visual camera's silicon sensor (which is sensitive to 850nm). ORB-SLAM3 uses these IR-illuminated frames for navigation — feature extraction works identically to daylight because the sensor receives sufficient photons for contrast detection. The IR illuminator consumes 2W and is visible to night vision goggles (NVG) within 500 meters — acceptable risk because NVG-equipped observers at 500m would also detect the drone acoustically.

Coordination Between Day and Night Shifts

Transitioning drone operations from daylight to darkness requires a structured handoff between day and night teams. At sunset minus 30 minutes: night team arrives at the drone station, receives situation briefing from day team leader, reviews Lisa 26 COP for current threat picture. Day team completes final daylight BDA pass. Night team switches all cameras to thermal-primary mode, verifies IR illuminator function on Fischer 26, and conducts a 5-minute calibration flight to confirm thermal detection is functioning. Day team departs. The 30-minute overlap ensures continuity of ISR coverage — no gap during the most tactically vulnerable period of fading light.

Try the interactive Pipeline Latency Analyzer →

Open the interactive Pipeline Analyzer →

← Del av Component Architecture

External source: Värmekamera – Wikipedia

Implementation

# Night Operations — Thermal Camera Detection
# pip install opencv-python
# pip install ultralytics
import cv2
from ultralytics import YOLO

# Load thermal-specific YOLOv8 model
model = YOLO("nordic_thermal_v2.engine")

# Open Infiray T2S+ thermal camera
thermal = cv2.VideoCapture("/dev/video1")
thermal.set(cv2.CAP_PROP_FRAME_WIDTH, 256)
thermal.set(cv2.CAP_PROP_FRAME_HEIGHT, 192)

def detect_thermal_targets():
    """Thermal detection loop — runs on Jetson Orin Nano."""
    while True:
        ret, frame = thermal.read()
        if not ret:
            continue
        
        # Lower confidence threshold at night (less visual noise)
        results = model(frame, conf=0.45, imgsz=256)
        
        for det in results[0].boxes:
            cls = int(det.cls)
            conf = float(det.conf)
            # Thermal: running engine 80-120C vs ambient 0-10C
            # Night mAP: 0.82 (vs day 0.75 — better in darkness)
            print(f"Detected: class={cls} conf={conf:.0%}")
            
    return results

# Night advantage: less visual noise, higher contrast
# Running engine = bright beacon on thermal at 2+ km

Related Chapters

Sources

ArduPilot documentation (ardupilot.org). Silvus Technologies documentation. NATO STANAG 4609 Ed. 4 (motion imagery metadata), STANAG 4671 (UAV airworthiness), and STANAG 2022 (intelligence source reliability). Specifically: Watling & Reynolds, "Meatgrinder: Russian Tactics", RUSI (2023); Bronk, Reynolds & Watling, "The Russian Air War and Ukrainian Requirements for Air Defence", RUSI (2022); ISW daily campaign assessments (understandingwar.org archive); CSIS Center for Strategic and International Studies Ukraine briefings. — FSG-A has no operational experience of its own. Swedish Armed Forces public documentation.