Automated Flight Planning
Precision mission control for repeatable results
1 min read
Algorithmic Precision
A deep look into how we turn raw video and GPS data into ICAO-compliant measurements with sub-pixel accuracy.
Published on Nov 19, 2025.
Last modified on Nov 19, 2025 at 12:00 am
The Processing Pipeline
From raw pixels to precise angles
1. Frame Extraction & Synchronization
The process begins with high-bitrate 4K video captured at 60fps. Crucially, we synchronize the video frames with the drone's RTK GPS logs using precise timestamps. This allows us to know the exact 3D position (Latitude, Longitude, Altitude) of the camera sensor for every single frame of video, down to the millisecond.
2. Sub-Pixel Light Detection
For each frame, we employ a custom blob detection algorithm. It isolates the PAPI lights based on high-intensity thresholds and chromaticity signatures. Instead of just finding the 'center pixel', we calculate the 'intensity centroid'—a weighted average of the pixel cluster. This technique allows us to determine the light's position on the sensor with sub-pixel accuracy (typically < 0.1 pixels), overcoming the resolution limits of the sensor.
3. Geometric Angle Calculation
We calculate exact vertical angles relative to two key reference points: the specific PAPI light unit and the runway touchdown point (aiming point). By triangulating the drone's GPS position against these fixed ground coordinates, we verify both the individual alignment of each light housing and the effective glide path angle that guides the pilot to the correct landing spot.
Accuracy Metrics
Theoretical Limits & Real-World Accuracy
Our system's accuracy is defined by two main factors: GPS positioning error and optical resolution.
- GPS Error Contribution.
- RTK GPS provides a vertical accuracy of ±1.5cm. At a measurement distance of 350m, this linear error translates to an angular error of just 0.0025° (approx. 0.15 minutes of arc).
- Optical Resolution Limit.
- Using a 4K sensor with a 24mm equivalent lens, one pixel spans approx. 0.0133° (0.8 minutes). Our sub-pixel centroiding improves this by 10x, reducing optical error to < 0.0017° (< 0.1 minutes).
- Gimbal Stability.
- Mechanical gimbal instability is mitigated by our algorithmic digital stabilization. This process mathematically corrects for vibration, but we still include the raw mechanical tolerance (±0.003°) in our error budget to ensure a conservative, fail-safe accuracy guarantee.
- Total System Accuracy.
- Summing these error sources ($0.0025° + 0.0017° + 0.003°$), our theoretical worst-case error is ±0.0072° (0.43 minutes). We conservatively guarantee ±0.017° (±1 minute), well within the ICAO tolerance of ±0.05°.
Method Comparison
How our drone-based method stacks up against traditional techniques.
| Method | Accuracy | Time / Runway | Repeatability | Disruption | Output |
|---|---|---|---|---|---|
| TarmacView Drone | Superior ±0.017° / ±1 min | ~20 mins | High Automated Path | Minimal No Closure | Digital Report Photo/Video Proof |
| Flight Inspection (Aircraft) | Standard ±0.05° / ±3 min | 2-4 Hours | Medium Pilot Dependent | High Runway Closure | Pass/Fail Analog Data |
| Ground Clinometer | Medium Static Only | 1-2 Hours | Low Human Reading | Medium Access Required | Manual Log No Visual Proof |
Accuracy Factors & Optimization
A detailed analysis of why each method performs the way it does and how accuracy can be maximized.
- TarmacView Drone
Our system achieves superior accuracy by completely eliminating human error. Unlike a pilot who must subjectively judge the color transition, our computer vision algorithms analyze the spectral signature of the light frame-by-frame. Combined with RTK GPS stability, this removes the variability of human reaction time and perception, resulting in consistent, objective measurements that exceed ICAO standards.
- Flight Inspection Aircraft
Traditional flight inspection relies heavily on the pilot's visual perception to identify the transition from red to white. This introduces significant human error, as the exact 'transition point' is subjective and can vary between pilots or even on different runs. Atmospheric turbulence and the aircraft's speed further degrade the precision of these subjective observations.
- Ground Clinometer
Clinometers measure the physical tilt of the light unit housing, not the actual light beam. While the device itself is precise (±1 min), it assumes the light bulb and lens are perfectly aligned with the housing. Accuracy is often compromised by 'settling' of the concrete foundation or human error in reading the bubble level. Digital clinometers and laser interferometry can reduce human error but cannot account for internal optical misalignments.
Unique Advantage
The Best of Both Worlds
Why choose between data and visual proof when you can have both? Our system is the only solution that combines the rigorous analytical precision of a laboratory instrument with the real-world visual verification of a flight check.
- Deep Analytical Data.
- We generate exact, high-accuracy numbers for every parameter. Our detailed charts visualize the intensity distribution, color spectrum, and angular transition with a level of granularity that analog methods simply cannot match.
- Visual Verification.
- Data isn't just abstract numbers. We provide the actual video frame corresponding to each measurement point. You can see exactly what the pilot sees, enhanced with overlay data, providing irrefutable optical confirmation of the results.
- Complete Validation.
- Other methods force a compromise: clinometers give you numbers without the visual context, and flight checks give you a visual check without precise numbers. TarmacView delivers both, ensuring total confidence in your PAPI system's safety.
Airport or Aviation Maintenance Professional?
We are building network of partners to revolutionize airport maintenance with cutting-edge technology.
Learn more
Technology & Innovation
State-of-the-art algorithms and secure cloud infrastructure
2 min read
Service Models
Flexible options for airports and opportunities for drone operators
2 min read