LWIR Guidance: Mastering IR Seeker for CUAS
Explore advanced strategies in LWIR guidance and IR seeker technology for CUAS. Learn how to maintain tracking under challenging conditions, manage occlusion, and improve target acquisition through model-driven techniques. Master the art of precision tracking with insights from expert practices.
10/9/20262 min read
At SpearX, building LWIR guidance (IR Seeker) for CUAS, we are learned one rule: losing the visual is normal. Losing the track is a choice.
Part 2: don't trust the brightest pixel. Part 4: trust the model when there is no pixel at all.
The target vanishes - cloud, building, smoke, or thermal crossover. A naive system widens the search and locks on noise. Fatal at 350 m/s closing speed.
What we do instead:
1. Hold the hypothesis, not the frame
Two vanishing types matter. Occlusion by an object, and occlusion by contrast - thermal crossover, where the target is still there but ΔT ≈ 0. The model holds in both; only the expected gap length differs.
2. Shrink the gate, do not widen it
Panic scans the whole FOV. Physics says the target sits inside a predicted uncertainty ellipse. We gate the ROI to that ellipse plus margin. Searching outside it only buys false locks.
3. Covariance is a countdown (coasting)
Dead reckoning is a clock, not magic. Without a measurement update, state covariance inflates every cycle. Position error grows quadratically when the target accelerates unseen, so the gate swells faster than linear intuition suggests. On uncooled LWIR our practical blind budget is ~300-600 ms before the ellipse stops paying rent.
4. IMU carries the platform, model carries the target
Gimbal and airframe motion are measured at 1 kHz. We subtract ego-motion so the predicted position stays in inertial space, not a drifting camera frame. This keeps the gate honest while the eye is closed.
5. Re-acquire by kinematics, not brightness
When the target pops back, the question is not "is it bright." It is "does this candidate sit in the predicted gate with a consistent velocity vector." Watch for ambiguity too: a second object can enter the same gate, so we re-check the last valid thermal texture, not just position, before resuming.
as a result of this:
Lost visual is not lost target. It is a switch from sensor-driven to model-driven tracking:
- hold the hypothesis through the gap;
- treat crossover as a contrast gap, not a lost target;
- gate the search to the uncertainty ellipse;
- treat covariance growth as a countdown;
- re-acquire by motion and texture, not by light.
next Part 5: the hardest occlusion is inside your own airframe - self-emission and thermal isolation.
At SpearX, building LWIR guidance (IR Seeker) for CUAS, we are learned one rule: losing the visual is normal. Losing the track is a choice.
Part 2: don't trust the brightest pixel. Part 4: trust the model when there is no pixel at all.
The target vanishes - cloud, building, smoke, or thermal crossover. A naive system widens the search and locks on noise. Fatal at 350 m/s closing speed.
Related : GEOCOM Co. LLC www.geocomco.eu DeepTechRnD www.deeptechrnd.eu SpearXAgro www.spearxagro.eu
Contacts
SpearX ￪ Strike with Precision
+359 877 620 210
D-U-N-S® 525540791
© 2025 - 2026. All rights reserved. SpearX Projects. Geocom Co. LLC
SpearX is an EU-Ukrainian company located in a strategic areas for R&D in UAV