Kira Vision — Methodology Deep-Dive

Where RTK Precision
Meets Thermal Inspection

We turned a $5k consumer drone into a bankable, IEC-compliant thermography platform.
The secret isn't hardware — it's where the photos land.

±2 cm RTK positional accuracy
2.3 cm/px Thermal GSD @ 17 m (M3T)
↓ Explore the methodology

The Industry Problem

Conventional Mapping Flights
Don't Follow the Modules

Every major player — Raptor Maps included — uses generic boustrophedon grid sweeps. The drone photographs the plant from random, inconsistent angles relative to module tilt.

⚠️

Angle Inconsistency

Variable viewing angles create specular reflections from sky and sun, corrupting radiometric readings at the sensor level.

🔥

Stitching Destroys Data

Thermal orthomosaics blend and blur raw radiometric pixels. The temperature you read is an interpolation artifact, not a measurement.

📉

False Positives at Scale

Reflections + stitching = phantom hotspots. O&M teams waste hours chasing anomalies that don't exist in the real data.

The RTK Solution

String-Aligned, RTK-Guided
Precision Flight Paths

01

RTK + DSM Terrain Follow

Flight plans generated from geospatial data: DSM-normalized altitude hold, string-aligned waypoints with RTK-fix precision. The drone follows the exact geometric line of each module row.

02

Constant Viewing Angle

Gimbal pitch is controlled to align with module tilt or maintain safe nadir (−90°). Every frame captures the panel's emitted radiation — not sky reflections. Repeatability across campaigns is guaranteed by geometry, not operator skill.

03

Proprietary KMZ Generation

No wasted photos. No idle hover. The camera fires only where modules exist. Optimized path routing (exact TSP for ≤16 rows) deposits overlap precisely on the target area — not on corridors and empty terrain.

Interactive Proof

Same Overlap. Same Cadence.
Different Deposition.

Both scenarios below use identical capture parameters: 2 m/s, 1 photo every 2s, 80% frontal overlap, same footprint. The only difference is where the photo shadows accumulate.

2.0 m/sCapture speed
1 photo / 2sSame interval
80% frontalSame overlap
Same footprintIdentical photo shadow
Speed 3.0×

Optimized flight (string-aligned)

Capture passes anchored to module rows; transitions happen off-target.

aligned to target

Conventional baseline

Boustrophedon sweep of the plant envelope with the same capture model.

reference baseline
Target rows (modules)
Total photo shadow
Useful overlap (on target)
Capture pass
Transit / connector

Key Takeaway

Calculating consolidated comparison…

Analysis Pipeline

Orthomosaic Is Reference.
Truth Lives in the Raw Photo.

The industry's biggest mistake: measuring temperatures on stitched orthomosaics. We do the opposite — the ortho is an index, the raw photo is the evidence.

STEP 1

RGB Orthomosaic

Global position reference. As-built extraction. Spatial inventory of every module with GPS coordinates and unique ID.

Reference only
STEP 2

Raw RGB Photo (LV2)

The original aerial photo at full resolution. Each cell individually resolved. Sub-pixel module segmentation via trained YOLO models.

High-resolution truth
STEP 3

Raw Thermal Photo (LV2)

Unaltered radiometric data. No stitching blur. Pure GSD at 2.3 cm/px (M3T @ 17 m, IFOV 1.33 mrad). The thermal audit and temperature extraction happen here — on the original, unmodified sensor output.

Bankable measurement

Visual Evidence

The Difference Is
Visible at First Glance

Commercial Methodology — GSD ~5 cm/px (M3T @ ~38 m), misaligned flight

Commercial RGB — high altitude, misaligned, half the frame is empty terrain RGB — High altitude, misaligned
Commercial Thermal — most of the frame captures irrelevant terrain Thermal — Mostly empty terrain

Kira Methodology — GSD 2.3 cm/px (M3T @ 17 m), string-aligned flight

Kira RGB — low altitude, perfectly aligned, every cell resolved RGB — Every cell individually resolved
Kira Thermal — modules fill the frame, hotspot clearly measurable Thermal — Hotspot clearly measurable

End-to-End Traceability: Module → RGB → Thermal → Ortho

Kira Report: Module #909 — thermal hotspot, high-res RGB crop, and orthomosaic position, all linked

Module #909 — Thermal anomaly (left), high-res RGB crop (center), orthomosaic position with bounding box and GPS (right). Everything linked. Everything traceable. CoA 3 classification.

User Experience

Click Any Module.
See the Real Photo.

The final deliverable is an interactive web report. The user clicks any module on the orthomosaic and instantly sees the original, unstitched RGB and thermal photos — not a blurry ortho pixel.

1

Open the plant orthomosaic — full overview

2

Click any module on the map

3

See the real RGB photo — each cell sharp and clear

4

See the real thermal photo — pure radiometric data

5

GPS coordinates, source photo ID, ΔT, CoA classification

Why This Only Works With RTK

Without centimeter-level georeferencing (RTK fix), you cannot project a module from the orthomosaic back to the raw photo with confidence. Without inter-sensor affine calibration, you cannot link the RGB pixel to the thermal pixel. The traditional orthomosaic is a dead end: beautiful to look at, but impossible to trace back to ground truth. Kira reverses the path: the ortho is the index, the raw photo is the evidence.

Hardware Disruption

$5,000 Drone Doing the Job
of a $25,000 Platform

SpecificationEnterprise (M300 + H20T)Mavic 3T + Kira
Platform cost~$25,000~$5,000
Weight9 kg (special license)920 g
Thermal sensor640 × 512640 × 512 (identical resolution)
IFOV0.89 mrad1.33 mrad
Max IEC altitude (≤3 cm/px)33.7 m22.5 m
Kira flight altitude—17 m (safety margin)
GSD at Kira altitude (17 m)1.5 cm/px2.3 cm/px
GSD margin vs IEC limit—24% below 3 cm/px
Analysis sourceStitched orthomosaicRaw photo (unaltered)
IEC complianceDepends on operatorBuilt into pipeline

Why IFOV Matters — and Why It's Not Enough

The H20T has a 1.5× better IFOV (0.89 vs 1.33 mrad), allowing it to fly at 33.7 m and still meet IEC's 3 cm/px limit. The M3T needs to fly at ≤22.5 m for the same GSD. But IFOV is a sensor property — it doesn't fix flight methodology. Most M300+H20T operators fly at 40-50 m (GSD 3.6-4.4 cm/px), exceeding IEC limits. With Kira's pipeline, the M3T at 17 m achieves 2.3 cm/px with a 24% safety margin — better than most H20T deployments in practice.

Summary

Kira vs. Industry Standard

AspectIndustry / Raptor MapsKira Vision
Flight alignmentGeneric grid (boustrophedon)String-aligned (RTK)
Viewing angleVariable, with reflectionsConstant, gimbal-controlled
Thermal GSD~5 cm/px (M3T @ 38 m typical)2.3 cm/px (M3T @ 17 m)
IFOV utilizationWasted — flies above IEC envelopeOptimized — 24% margin below 3 cm/px
Analysis sourceStitched orthomosaicOriginal raw photo
TraceabilityModule on map, no photo linkModule → RGB → Thermal
Flight efficiencyPhotos wasted on empty terrainEvery photo on target
IEC compliancePartial (GSD > 3 cm/px at typical altitude)Full (17 m, GSD 2.3 cm/px, ≥75% overlap)
False positivesFrequent (reflections, stitching)Minimized (raw data, constant angle)