Air Pixels
Living document · Rev 1.0

Air Pixels · Research & IP register

Patents & proposed experiments.

Air Pixels builds flight software, provenance tools and drone data products. This register lists the inventions we plan to protect, where each one stands in the patent process, and the experiments we'll run to test what we say about our work. We update it whenever a status changes.

Read the registerSee the experiments
  1. 01Disclosure

    Written up, dated and signed by the inventors. Kept internal.

  2. 02Search

    Prior art and freedom to operate reviewed. Route chosen: patent, trade secret or open.

  3. 03Provisional

    Filed with the USPTO. Patent pending starts here, with 12 months to file in full.

  4. 04Full application

    Non-provisional with claims, under examination. International filing if we need it.

  5. 05Decision

    Granted with a patent number, or closed with the reason.

§ 1

Invention register.

Six inventions from Proof of Flight, AI & Authenticity, Scan-to-BIM, the Drone Comparison Engine and Flight OS. Abstracts repeat only what's already public on airpixels.com. One entry is withheld until it's filed.

Table 1 · Register at a glance
IDInventionRouteStageFirst public disclosure
AP-INV-001Proof of Flight decision recordsPatent01 · in progress2026-10-27 · /framework
AP-INV-002Track match verificationPatent01 · in progress2026-10-27 · /authenticity
AP-INV-003Flight assertion schemaOpenDraft schema2026-10-27 · /authenticity
AP-INV-004Breakline-preserving terrain reductionPatent, if searched clear01 · in progress2026-10 · /scan-to-bim
AP-INV-005Field-adjusted endurance modelTrade secretNo filing2026-10 · /drones, method only
AP-INV-006Withheld until filingPatent01 · in progressNot disclosed

AP-INV-001 · Proof of Flight

Patent routeUS grace period · 365 days left

(54)Tamper-evident recording and multi-party verification of decisions made by autonomous aircraft

  1. DisclosureIn progress
  2. Search
  3. Provisional
  4. Full application
  5. Decision

(57)Abstract

An autonomous aircraft writes a record for each decision it makes: a digest of the inputs it saw, the build hash of the model that proposed the action, the options it ranked, the rule checks that passed or vetoed it, and reason codes. Each record carries the hash of the one before it. Batches are sealed into Merkle roots that several independent parties co-sign, and the ledger is checkpointed to a public chain, so a single delivery can be verified from a short proof.

(21)Application
Not filed
(52)CPC, indicative
B64U · G06F 21/64 · H04L 9/50 · G06N 5/045
Disclosed
2026-10-27 · /framework
Grace period ends
2027-10-27 [2]
Related work
Hash-chain timestamping [15], Merkle trees [16], Certificate Transparency [17], crash-protected flight recorders [18]
Tested by
AP-EXP-02
Next step
Prior-art search, then a provisional before any further technical detail is published.

AP-INV-002 · Track match

Patent routeUS grace period · 365 days left

(54)Verifying aerial video against the logged track of the flight it is attributed to

  1. DisclosureIn progress
  2. Search
  3. Provisional
  4. Full application
  5. Decision

(57)Abstract

Camera positions are solved for every frame of a clip with structure-from-motion, aligned to the flight log the clip is attributed to, and compared point by point. The deviation between the two, checked alongside timing, altitude and airspace, shows whether the clip could have been recorded on that flight. A generated clip has no log to match, and a forged log has to agree with the footage, the aircraft and the airspace authorization at once.

(21)Application
Not filed
(52)CPC, indicative
G06T 7/73 · G06V 20/17 · G06F 21/64 · B64U
Disclosed
2026-10-27 · /authenticity
Grace period ends
2027-10-27 [2]
Related work
Structure-from-motion [19], least-squares point-set alignment [20], Content Credentials [21]
Tested by
AP-EXP-01
Next step
Prior-art search, then a provisional.

AP-INV-003 · Flight assertion

Open · free to use

(54)A Content Credentials assertion that binds media to the flight that captured it

  1. Draft schemaIn progress
  2. Public comment
  3. v1.0 published

(57)Abstract

A signed assertion in a file's Content Credentials records the aircraft serial, takeoff time, a hash of the flight track and the airspace authorization behind a clip. We're publishing it as an open schema for operators and tool makers to use and critique, so this entry is a defensive publication. We won't patent it.

(21)Application
None, open by design
Field
Media provenance
Disclosed
2026-10-27 · /authenticity
Related work
C2PA specification [21]
Next step
Publish schema v0.1 for public comment.

AP-INV-004 · Scan-to-BIM terrain & levels

Patent, if searched clearUS grace period · ends Oct 2027

(54)Breakline-preserving terrain reduction and automatic datum levels from aerial point clouds

  1. DisclosureIn progress
  2. Search
  3. Provisional
  4. Full application
  5. Decision

(57)Abstract

Ground points from a classified drone point cloud are reduced to a vertex budget a BIM tool can load, holding elevation error under a set tolerance while keeping retaining walls, curbs and ditches as sharp breaklines. Levels such as grade, eave and ridge are set from the measured building geometry and exported with the terrain as IFC 4.3.

(21)Application
Not filed
(52)CPC, indicative
G06T 17/05 · G06T 17/20 · G06F 30/13
Disclosed
2026-10 · /scan-to-bim
Grace period ends
2027-10 [2]
Related work
Cloth simulation ground filtering [22], quadric error simplification [23], IFC 4.3 [24]
Tested by
AP-EXP-04
Next step
Prior-art search. If close prior art turns up, the implementation stays a trade secret instead.

AP-INV-005 · Field-adjusted endurance

Trade secret

(54)Predicting usable flight time from rated specifications, flight conditions and logged flights

No filing. A trade secret stays protected for as long as it stays secret, with no fixed term.

(57)Abstract

Rated flight and hover times are reduced for wind against the aircraft's rated limit, battery temperature, thinner air at elevation and the pilot's landing reserve. The method summary is public on the Drone Comparison Engine. The coefficients behind it, fitted to logged flights, stay internal.

(21)Application
None, trade secret
Field
Flight performance
Disclosed
Method only · 2026-10 · /drones
Why not a patent
A formula on its own is hard to patent [10] and hard to detect in someone else's software. The value is in the calibration.
Controls
Coefficients and training logs on a need-to-know basis [13]
Tested by
AP-EXP-03

AP-INV-006 · Flight OS

Patent routeWithheld

(54)Title withheld until filing

  1. DisclosureIn progress
  2. Search
  3. Provisional
  4. Full application
  5. Decision

(57)Abstract

Withheld until a provisional is on file
(21)Application
Not filed
Field
Flight OS · mission planning
Disclosed
Not disclosed
Why withheld
Publishing first would start the US clock and could rule out a patent in Europe, China and many other countries. [2]
Next step
Prior-art search, then a provisional.

§ 2

Proposed experiments.

Each one tests a claim we've already made in public. Hypotheses, samples and pass criteria are fixed here before any data is collected. Results are posted whether they pass or fail.

  1. AP-EXP-01

    Protocol v0.1 · not run

    Does a solved camera path land on its flight log?

    For a flown shot, that solved path lands on the flight log within a few meters.
    Claim tested · AI & Authenticity

    Linked · AP-INV-002

    Hypothesis
    Camera paths solved from flown clips sit within 5 m of their logged track. Logs from other flights don't match.
    Design
    120 clips flown by Air Pixels on 3 airframes, 30 to 120 m AGL, over urban, suburban, rural and water scenes. Poses solved with COLMAP [19], aligned to each log with a similarity transform [20] and compared frame by frame.
    Controls
    Every clip is also scored against logs from other flights at the same site. Repeat runs of the same waypoint mission, the hardest case, are reported separately.
    Primary metric
    Mean deviation per clip, in meters
    Starts
    After the AP-INV-002 provisional is filed
    Limits
    Clips edited after capture, such as speed ramps and crops, are left for a follow-up study.

    Pass criterion · fixed before the run

    At least 115 of 120 flown clips match at 5.0 m or less (lower 95% bound ≥ 90%), and under 1% of wrong-log pairs do.

  2. AP-EXP-02

    Protocol v0.1 · not run

    Can an aircraft keep its own record without slowing down?

    Anyone with access can check a flight. No one can quietly rewrite one.
    Claim tested · Proof of Flight

    Linked · AP-INV-001

    Hypothesis
    Writing a signed record adds no more than 2 ms to each decision, no record is lost to a dropped link, and every edit made after anchoring is caught.
    Design
    500 simulated delivery flights in PX4 software-in-the-loop with random link outages of up to 10 minutes, plus bench timing on a Raspberry Pi 5-class companion computer. 10,000 tampering attempts: edits, deletions, reordering, insertions and full re-hashing.
    Primary metrics
    Record-write latency at p99 · share of records anchored · share of tampering detected
    Starts
    After the AP-INV-001 provisional is filed
    Limits
    A compromised aircraft that records false inputs at the moment of decision can't be caught this way. Sensor attestation is a separate study.

    Pass criterion · fixed before the run

    p99 latency ≤ 2 ms, 100% of records anchored after reconnect with every outage logged, and 100% of post-anchor tampering detected.

  3. AP-EXP-03

    Protocol v0.1 · not run

    How close are field-adjusted flight times to real flights?

    Drone specs, recalculated for the field.
    Claim tested · Drone Comparison Engine

    Linked · AP-INV-005

    Hypothesis
    Predicted usable flight time lands within 10% of measured time to landing reserve, and closer than the rated figure on most flights.
    Design
    150 logged flights across at least 6 aircraft models, from −5 to 35 °C, in winds of 0 to 25 kt, from launch elevations up to 2,500 m. Logs come from our fleet and from pilots who opt in, with personal data removed.
    Baseline
    The manufacturer's rated flight time
    Primary metric
    Mean absolute percentage error against measured time to reserve
    Starts
    Once 150 logs are collected
    Limits
    Payloads and third-party batteries are excluded. Results report error only. The coefficients stay internal.

    Pass criterion · fixed before the run

    Error of 10% or less, mean bias within ±3%, and closer than the rated figure on at least 80% of flights.

  4. AP-EXP-04

    Protocol v0.1 · not run

    Does reduced terrain keep its shape?

    Flat lawns are reduced to minimal vertices within 1 inch of elevation error, while retaining walls, curbs and ditches are kept sharp.
    Claim tested · Scan-to-BIM

    Linked · AP-INV-004

    Hypothesis
    Reduced terrain stays within 1 inch RMSE of the full-density ground surface on every site, with breaklines intact.
    Design
    12 sites across flat lots, hillsides and dense urban parcels, each flown with LiDAR or photogrammetry and checked with at least 30 RTK GNSS shots, following ASPRS Edition 2 [25].
    Primary metric
    Vertical RMSE of the reduced surface against full-density ground points
    Secondary
    Absolute vertical accuracy against check shots · breakline position error
    Starts
    After the AP-INV-004 prior-art search
    Limits
    Absolute accuracy depends on capture and ground control, so it's reported but not scored.

    Pass criterion · fixed before the run

    RMSE ≤ 1.0 in (25.4 mm) on all 12 sites, under 15,000 vertices, and breaklines within 2 in horizontally.

  5. AP-EXP-05

    Protocol v0.1 · not run

    Does the Ground Truth check help people spot generated aerials?

    The Ground Truth check: 16 observations across provenance, physics and pixels.
    Method tested · AI & Authenticity

    Linked · Open method, no patent

    Hypothesis
    Reviewers using the check sort flown from generated clips more accurately than reviewers without it, and don't accuse more real footage.
    Design
    200 clips: 100 flown by Air Pixels and 100 generated by four current video models from prompts matched to the flown scenes. 40 reviewers, randomly assigned to use the check or not, each rate 50 clips blind. Clips carry no metadata, so only the physics and pixel layers are tested.
    Primary metric
    Balanced accuracy
    Secondary
    Share of flown clips called generated
    Starts
    No filing dependency
    Limits
    Video models change quickly. Results are dated to the models tested, and the study will be rerun.

    Pass criterion · fixed before the run

    The check group is at least 10 points more accurate, with 5% or fewer flown clips called generated.

§ 3

From disclosure to decision.

The legal clock for a US application, and the rules we hold ourselves to along the way.

Fig. 1 · A US patent application, month by month

Months counted from the provisional filing (M0). Step 1 applies only to inventions published before filing, as AP-INV-001, 002 and 004 were.

  1. 1 · M−12

    Public disclosure

    If we publish before filing, US law gives us a year to file. In Europe, China and many other countries, publishing first can rule out a patent. [2]

  2. 2 · M0

    Provisional filed

    Patent pending begins. The provisional isn't examined and lapses after 12 months. [3]

  3. 3 · M12

    Full application

    A non-provisional with claims, plus a PCT application to keep other countries open. [3] [4]

  4. 4 · M18

    Publication

    The application is published 18 months after the earliest filing date. [5]

  5. 5 · M30

    National phase

    The PCT application enters each country we choose. [7]

  6. 6 · M36+

    Decision

    Examination often takes two to three years, and Track One aims for about 12 months. A patent runs 20 years from the full application's filing date. [8] [9]

Rules we follow

  1. "Patent pending" only with an application number.

    Marking something patent pending with nothing on file can be false marking under US law. [1] Each entry shows its application number once it has one.

  2. File first, then publish.

    Proof of Flight, track match and Scan-to-BIM went public before this register existed, so their US clock is already running. From here on, details of unfiled inventions stay internal until a provisional is on file.

  3. Pass criteria come before results.

    Each experiment's hypothesis, sample and pass criterion are published before data collection starts. Results are posted whether they pass or fail.

  4. Inventors are people.

    Only people can be named as inventors. [6] We use AI tools in research and engineering, and our records note who conceived each idea and which tools helped.

  5. Open formats stay open.

    We won't patent the Flown, Assisted and Generated labels, the Ground Truth check or the flight assertion schema. Publishing them with a date keeps them free for anyone to use.

  6. Every change is logged.

    Filings, application numbers, results and dropped entries go into the record with a date.

§ 4

Beyond patents.

Most of what we make is better protected another way, or kept open on purpose.

Table 2 · Protection map
WhatProtectionHow we handle it
Inventions in the registerPatentProvisional first, then a full application within 12 months. Status per entry in § 1.
AIR PIXELS name and pixel-V markAlso PROOF OF FLIGHTTrademarkClearance search, then USPTO applications. Shown with ™ until registered, and with ® only after. [11]
Writing, design, code and downloadsMove Book, LUTs, presets, templatesCopyright© notice on every page and file. Key works registered before we'd ever need to enforce them. [12]
Drone Shot Database and Delivery TrackerCopyright in the compilationThe facts are free to cite. The selection and arrangement are ours, so please link back when you use them. [14]
Model weights, calibration data, scoring thresholds, unfiled inventionsTrade secretNeed-to-know access and confidentiality agreements. Never published here. [13]
Labels, Ground Truth check, flight assertion schema, reason-code catalogueOpenLabels and check published, schema in draft, catalogue planned. Each is dated when published so it stays free for anyone to use.

§ 5

Revision record.

Public disclosures, filings, results and changes of route, each with its date. Newest first.

Table 3 · Record
DateChangeEntries
2026-10-27Register opened at Rev 1.0 with six inventions and five proposed experiments. No applications on file.All
2026-10-27AI & Authenticity published at /authenticity. First public disclosure of track match and the flight assertion.AP-INV-002, 003
2026-10-27Proof of Flight published at /framework and /framework/ledger. First public disclosure.AP-INV-001
2026-10Scan-to-BIM engine published at /scan-to-bim.AP-INV-004
2026-10Drone Comparison Engine method summary published at /drones.AP-INV-005

§ 6

References.

Statutes, cases, standards and papers cited on this page.

Law & procedure

  1. [1]
    35 U.S.C. § 292. False marking.
  2. [2]
    35 U.S.C. § 102(b)(1). Disclosures made one year or less before the effective filing date.
  3. [3]
    35 U.S.C. §§ 111(b), 119(e). Provisional applications and the benefit of their filing date.
  4. [4]
    Paris Convention for the Protection of Industrial Property, Art. 4. Right of priority.
  5. [5]
    35 U.S.C. § 122(b). Publication of applications.
  6. [6]
    Thaler v. Vidal, 43 F.4th 1207 (Fed. Cir. 2022). Inventors must be natural persons.
  7. [7]
    Patent Cooperation Treaty, Art. 22. National phase at 30 months from priority.
  8. [8]
  9. [9]
    USPTO. Prioritized Patent Examination Program (Track One).
  10. [10]
    Alice Corp. v. CLS Bank International, 573 U.S. 208 (2014). Patent eligibility of abstract ideas.
  11. [11]
    15 U.S.C. § 1111. Notice of registration.
  12. [12]
    17 U.S.C. § 411(a). Registration before an infringement action.
  13. [13]
    18 U.S.C. § 1839(3). Definition of a trade secret.
  14. [14]
    Feist Publications, Inc. v. Rural Telephone Service Co., 499 U.S. 340 (1991). Facts and compilations.

Technical

  1. [15]
    Haber, S. and Stornetta, W. S. (1991). How to time-stamp a digital document. Journal of Cryptology 3(2), 99–111.
  2. [16]
    Merkle, R. C. (1988). A digital signature based on a conventional encryption function. CRYPTO '87, LNCS 293, 369–378.
  3. [17]
    Laurie, B., Langley, A. and Kasper, E. (2013). Certificate Transparency. RFC 6962, IETF.
  4. [18]
    EUROCAE ED-112A (2013). Minimum operational performance specification for crash protected airborne recorder systems.
  5. [19]
    Schönberger, J. L. and Frahm, J.-M. (2016). Structure-from-motion revisited. CVPR. Implemented in COLMAP.
  6. [20]
    Umeyama, S. (1991). Least-squares estimation of transformation parameters between two point patterns. IEEE TPAMI 13(4), 376–380.
  7. [21]
    Coalition for Content Provenance and Authenticity. C2PA technical specification.
  8. [22]
    Zhang, W. et al. (2016). An easy-to-use airborne LiDAR data filtering method based on cloth simulation. Remote Sensing 8(6), 501.
  9. [23]
    Garland, M. and Heckbert, P. S. (1997). Surface simplification using quadric error metrics. SIGGRAPH '97, 209–216.
  10. [24]
    ISO 16739-1:2024. Industry Foundation Classes (IFC 4.3) for data sharing in the construction and facility management industries.
  11. [25]
    ASPRS (2023). Positional Accuracy Standards for Digital Geospatial Data, Edition 2.

Prior art & participation

Know of earlier work?

If you know of earlier work that covers one of these inventions, send it with a link and its date. We'd rather know before we file. Pilots who'd like to contribute flight logs to AP-EXP-03 can write to the same address. Personal data is removed before analysis.

Send prior artContribute flight logs
AP-RD-001 · Rev 1.0 · Oct 27, 2026 · © 2026 Air Pixels, Inc.This register describes plans and status. It isn't legal advice or a license, and it doesn't claim that any patent exists.