How Omniris turns what a drone survey finds into work items on a map: placed where they are on the ground, filed once, worked from a phone on site, shared with the people who need them, connected to your other systems, and kept in the same tamper-evident record as the imagery they came from.
FindingF-000001 · Tarp on the access road
SourceAI detection · tarp · 91% confidence
Fromone angled photograph · 4000 × 3000
Placedprojected onto the ground from the camera · ±30 ft (9.1 m)
Evidenceimage cut from the photograph · fingerprints recorded
FiledSite A · Roof damage · Medium
Historycreated · assigned · In progress · comment from the field
Recordevery step sealed in the custody record
Illustrative example values.
Summary
A drone survey is flown to find something: a hot connector, a tree growing into a power line, a washout, a damaged roof, a change since last month. What matters is the list of what was found, and what happens to each item next. In Omniris each item becomes a finding: a work item with a place on the map, a category, a severity, a status, a person responsible, evidence images and a complete history.
One list for everything found
Drawn by hand, found by AI, by a thermal analysis, by a LiDAR clearance check or a change alert, or reported from a phone on site: one list, one map, one set of filters.
On the ground, not in a photo
Something spotted in an angled photograph is placed where it is on the ground, with a stated margin of error.
Never filed twice
A tarp seen in six overlapping photos becomes one finding with six pieces of evidence when their positions agree. Next month's flight adds to the open finding instead of opening a new one.
Worked in the field
The mobile app lists what is assigned to you, guides you there, and lets you change the status, comment, add photos and report new findings, with or without signal.
Shared and connected
Findings travel on share links and in emails, reach your other software automatically, and export to GIS, spreadsheets and Google Earth.
Signed, held, provable
Every change is sealed in a tamper-evident record. A signed certificate shows who did each step, and a legal hold stops a finding and its evidence from being deleted.
Why findings get lost
Most inspection programmes do not fail at capture. They fail between the survey and the repair:
Detections are not work. AI draws boxes on photographs; a thermal analysis marks hot spots on a map. Neither says who will look at it, by when, or whether it was dealt with.
The same problem is reported many times. Overlapping photographs show it from several angles, and every repeat flight finds it again. A list that counts each sighting as a new problem buries the real ones.
A box on a photo is not a place. A crew needs a spot on the ground. A box on an angled photo doesn't say where to walk unless you also know where the camera was, where it was pointing and the shape of the ground.
The record is somewhere else. The work order lives in a spreadsheet or another system, cut off from the imagery that justified it, with no proof of who changed what and when.
Omniris keeps the finding, its evidence and its history together, in the same vault and under the same tamper-evident custody record as the survey that produced it.
From detection to work item
A finding can be made by hand at any time. Results from an analysis become findings in one of two ways. A person promotes a result they are looking at (turns it into a finding with one click). Or the analysis is started, or the change alert is set up, with Create findings on, and its results become findings automatically when it finishes. An AI job can instead wait for a person: its findings are then made image by image, as someone checks and verifies each image's detections.
flowchart LR
A[AI detections thermal hotspots LiDAR clearance change alerts AI change annotations] --> B[Placed on the ground]
B --> C[Grouped: one per thing]
C --> D[Matched to open findings]
D --> E[Category and severity by rule]
E --> F[Evidence images cut and sealed]
F --> G[Assigned, notified, worked]
1
Placed on the ground
Results on a map, such as an orthomosaic (a stitched, map-accurate aerial image), are already in the right place. Something seen in a single photograph is traced from the camera's position and angle to the ground, using the site's 3D surface model when there is one, with a margin of error that says how precisely the place is known.
2
Grouped
Sightings of the same thing in the same run become one finding with several pieces of evidence, the most confident one first.
3
Matched
A sighting that falls on an open finding of the same kind is added to it as new evidence, noted in its history, instead of opening a second finding.
4
Categorised by rule
Each organization or vault sets rules that give a result a category and severity: by AI label, by how much hotter or colder than its surroundings, by how close vegetation is to a line. Whoever starts an AI job or promotes results can choose differently for that one time.
5
Evidence
An image of what was found is cut from the source photograph or map, and its digital fingerprint is sealed into the record. If the source file is encrypted, so is the evidence image.
6
Assigned and notified
The person assigned gets an email; an automatic run sends one summary email, not one per finding. Your other software can be told automatically.
An automatic run finishes on the server even if the person who started it closes the browser, and a run that is retried never creates a finding twice. By default an automatic run files at most 500 findings, the most severe first; the organization's managers or the vault's owner can change that limit, and the rest can be promoted in one step.
Where findings come from
Source
Becomes a finding when
Its title says
By hand
Drawn on the map, on a photograph or orthomosaic, or clicked onto a 3D model; a shape drawn on a photograph is placed on the ground for you
What the person wrote
On site, from a phone
Reported in the mobile app, with a photo if you take one, at the phone's position and the GPS accuracy the phone reports
What the person wrote
AI detection
Promoted from the AI results on an image, or created by an AI job: when the job finishes, or as a person verifies each of its images
"Tarp detected"
Thermal hot or cold spot
Promoted from the thermal results, or created by the analysis; the value is the largest difference from the surroundings
"Hot spot 22.7 °F (12.6 °C) from background"
LiDAR clearance
Promoted from, or created by, the vegetation clearance check of a LiDAR job along a power line
"Critical: vegetation 3.9 ft (1.2 m) from conductor"
Change alert
Each area that changed beyond a site's alert thresholds, created automatically when the alert has Create findings on, or promoted from the alert's results; with an image of the area when an orthomosaic covers it
The change, with its size and direction
AI change
Two map images of the same site from two dates, both analysed by the same AI model, are compared in the Site Navigator. Every object that appeared or disappeared becomes a new finding automatically; objects seen on both dates are not filed. It is filed under Other with no severity until the organization or vault adds a rule for it, and its evidence image shows both dates side by side
"Car appeared" or "Car disappeared"
Annotation
An existing annotation (a note pinned on an image) is made into a finding, with a category chosen at that moment
The annotation's first note line
Titles and values are written in feet and degrees Fahrenheit first, with metric alongside. A finding keeps a link to what it came from: the detection, the thermal analysis, the LiDAR job, the alert, the AI change comparison or the annotation.
Placing a finding on the ground
Most detections are found in single photographs, many of them taken at an angle. Omniris traces each one from where the drone was, which way the camera pointed, the lens and the height above the ground, onto the site's newest 3D surface model when there is one, otherwise onto the ground beneath the drone.
Every placement carries an uncertainty radius: a circle that should contain the true spot 95% of the time. It allows for how precise the drone's position was (whether it had an RTK fix, a centimetre-level satellite position), how accurately the camera's angle is known, how steeply it looked at the ground and how well the ground height is known. When the camera looked close to the horizon, or the photo has no record of where the camera pointed, the finding is placed at the drone's own position with a radius wide enough to say so honestly. A photo with no position at all is refused, not guessed.
How the radius is set
Without an RTK fix, the radius assumes a 16 ft (5 m) position error. A camera whose lens is not yet in Omniris' lens catalogue is placed with a generic lens and a radius widened by half to allow for it; cameras in the catalogue use their own published lens data.
Never the same problem twice
Overlapping photos are what make a survey reliable, and what make a simple list useless. Omniris combines sightings with rules that are cautious by design:
Within a run, sightings are combined only when they are the same kind (the same AI label, the same category) and close enough that their uncertainty circles overlap. A group cannot creep along a row of objects: it may not grow wider than twice its largest radius.
Two sightings in the same photograph are two things, never one, except where the AI split one object across the edge between two parts of the image it analysed separately.
Against existing findings, a sighting is added to the nearest open finding of the same kind it overlaps. Resolved and dismissed findings are never reopened by a new sighting.
Imprecise positions are never combined. A sighting whose uncertainty is over 82 ft (25 m) always stands alone.
A change in the opposite direction is a new finding. Ground lowered where it had been raised is a different event, not more evidence for the first one.
Every grouping and match is visible: a finding shows how many sightings it was built from, and its history records each match.
When a detection is checked and corrected
A person can check AI detections, correct them where the AI was wrong, and mark the image Verified, with a sealed record of who verified it and what they changed. A finding built on a detection does not go on showing what the AI got wrong.
Verified results come first. Once an image is verified, every finding made from it uses the corrected boxes, outlines and labels. Its evidence image shows the person's outline, not the AI's.
Findings only from checked results. An AI job can be started so that nothing becomes a finding until a person has verified the image. This applies to the images the job ran on; a copy of them, including one delivered to a client, never creates findings by itself.
A correction asks about the findings it touches. When a detection that is already evidence on a finding is moved, resized, relabelled or deleted, the person verifying is shown the findings they are allowed to see and decides for each one: false positive, or keep.
A kept finding follows the correction. Its evidence image is cut again from the corrected box. Its position is worked out again, from the corrected box, or from the corrected outline when only the outline was redrawn. A corrected label is carried over. The earlier evidence image is kept, and its fingerprint is sealed into the history. Evidence of a deleted detection stays on the finding, marked as removed, with the verification and the person who removed it.
A position a person set is never overwritten. A finding moves with a corrected box only when its position was worked out from that detection. If someone placed it by hand, the history says the location was kept, and who set it.
A finding never blocks a verification. A finding under a legal hold, one already closed, or one changed by someone else in the meantime is left as it is and listed with the reason. The verification is recorded regardless.
A correction never passes unnoticed. When nobody decides about a finding (the person verifying cannot see it, or chose to leave it), the finding is not changed, but its history records that the detection behind its evidence was changed or removed, and in which verification. It does not say what the correction was or who made it: someone who can read the finding may not be allowed to open the image.
Nothing fails silently. If a kept finding's new position cannot be worked out after repeated attempts, the finding keeps the position it had, and its history says so.
Each of these steps is an entry in the finding's history and in the custody record, and reaches your connected software like any other change. People are shown by name, never by email address.
Working a finding
Every finding moves through four stages. Each organization or vault sets its own statuses within each stage, its categories and its severities, starting from these defaults:
The Findings view lists and maps every finding you can see. Filter by stage, status, severity, category, assignee and site, or search by title, description or reference number, and update a whole selection at once.
Groups and the folder filter. Findings can be put into named, coloured groups of your own, such as "North roof" or "Work order 12"; a finding can be in any number of groups, and each change is recorded in its history. Anyone who can work findings in the organization or vault can create, rename, recolour, archive or delete its groups, except that a pilot can do so only in their own vault or an organization they manage. Anyone who can work a finding can add it to a group or take it out. The list, the map and the exports can be narrowed to one group, or to one folder and every folder below it that you can see; exports list each finding's groups.
A finding belongs to a folder. Whoever can see the folder can see the finding; whoever can change files in it can work the finding. The person assigned can work their own open finding (status, evidence, comments) even if they can only view the folder.
Assignment is to a person; they are emailed, and the count of their open findings is shown on the web and in the mobile app.
Comments and evidence can be added by anyone who can work the finding, and by the person assigned. An author can edit or delete their comment for 15 minutes; the history records that it was deleted.
The Site Navigator (the site map view) shows a site's findings beside its maps and models, with "New finding here".
In the field
The Omniris mobile app is where findings get resolved. Its Findings tab shows a badge counting the open findings assigned to you, and lists findings or shows them on a map around you.
Everything about the finding: its evidence images, measurements, location and uncertainty, history and comments, and Navigate to open the phone's maps app at the spot.
Work it on the spot: change the status or the person assigned, comment, and add photos from the camera. A person assigned who can only view the folder can still add photos; Omniris files them in the finding's folder.
Report something new: file a finding at the phone's position, with the accuracy the phone reports and a photo if you take one, in a folder you choose.
Promote a detection from the AI results on a photo in the app.
No signal, no problem. Changes made offline wait in a visible queue and are sent when the phone is back online. The 25 most recently updated open findings assigned to you are saved on the phone so they open with no signal.
Edits never silently overwrite. If someone changed a finding while you were offline, the app shows what changed and lets you keep their version or apply yours.
A shared phone is safe. Signing out removes the person's saved data from the phone. Changes one person queued are never sent under another person's sign-in; they wait for that person to sign back in.
Updating many findings at once, settings and exports are on the web and the desktop application; the phone is for the work itself.
In the viewers and on shared links
Image viewer. Findings appear on the photographs and orthomosaics they were found in. A finding can be drawn directly on an image, as a pin or an outline, and Omniris places it on the ground.
3D viewer. Findings appear as markers in the model, and a new one can be placed by clicking the surface.
Panoramas. Findings within 3,281 ft (1,000 m) appear as labels in their direction from the camera.
The desktop application does all of the above, the same way.
Share links. A public link to a folder, a 3D model or a Site Navigator view can include its findings; the box is unticked by default. Recipients can read but not change them. They see the complete record, including the names of the people involved, but never an email address. A link to a 3D model or a Site Navigator view also shows the findings filed on the flights it was built from, and a recipient can request a signed certificate through the link.
Connected to your systems
Channel
What it does
Email
"Assigned to you", and one summary per automatic run. Sent under the System notifications email setting; one click unsubscribes (which also stops job emails).
Webhooks
Automatic, signed messages to your own software whenever a finding is created, changed, assigned, given evidence or deleted (comments are not sent). You choose which changes and which findings. A bulk change arrives as one message, split into parts of 500 findings when it is larger. A failed delivery is tried again, waiting longer each time up to six hours between tries, up to 12 attempts in all, and every delivery is logged for 30 days, with Redeliver and Send test event. Up to 25 per organization or vault.
API
Access keys (tokens) per organization or vault, read-only or read and write, let your own software list, filter, fetch and export findings; change status, severity, category and assignee; comment; and add evidence. A change feed, including deletions, keeps another system in step. Changes are safe to resend.
Exports
GeoJSON for GIS, CSV for spreadsheets, KML for Google Earth: every finding matching the filters, up to 20,000.
Set up by the people in charge. Webhooks and API tokens are managed by an organization's managers or a personal vault's owner. They switch off by themselves if the person who created them loses that role or their account.
Attributed. A change made through a token appears in the finding's history under the token's name, and the custody record names the person who created it.
Webhooks only go where they should. Addresses must be public, secure (HTTPS) web addresses; addresses inside the hosting network are refused, and redirects are never followed. An on-premises site can allow named addresses on its own network.
Nothing secret in a webhook. A delivery carries the finding's details and the fingerprints of its evidence, never a link to a file, a storage location or a password.
Evidence you can prove
A finding is a claim that something on a site needed attention. Omniris records it with the same tamper-evident custody as the imagery behind it:
Every change is sealed. Creating, editing, assigning, commenting, adding evidence, deleting and restoring each become an entry in the custody record, naming who did it: a person, an automatic run or an API token.
Finding certificates. A signed PDF lists the finding's complete history with who did each step, the fingerprints of its evidence and the finding as that history describes it. It carries its own proof, can be checked online or offline, and proves that no step of the history was left out.
Legal holds and retention. While a finding's folder or any of its evidence files is under a legal hold or a retention policy, the finding cannot be deleted or lose evidence. Placing a hold on a folder brings its deleted findings back from the recycle bin.
Recycle bin. A deleted finding goes to the recycle bin with the rest of the vault and can be restored until it is permanently deleted.
Deliveries. When a pilot delivers a job's results to a client, the findings filed in the delivered folders go with them, with their history, comments and evidence from the delivered files.
People who leave. When an account is deleted, its name is erased from finding history, and later certificates show "a person (name erased)".
The companion paper, Evidence you can prove, describes the custody record, certificates, holds and retention in full.
Price
Findings, webhooks and API tokens are included. There is no charge to create, promote, work, share or export findings. Evidence images count toward the organization's or vault's storage, like any other file. The analyses that produce detections (AI, thermal, LiDAR, change detection) are priced as they always are.
Where it runs
Findings are part of Omniris wherever it runs: in the cloud on Microsoft Azure or Google Cloud, in the data center edition, and on-premises. Every change to a finding is sealed in the tamper-evident custody record.
What a finding claims, and what it doesn't
A finding is a work item, not a determination. It records that something was found and what was done about it. Whether it breaches a standard is for a qualified person. A LiDAR clearance result is an as-flown measurement, not a compliance determination, and so is a finding made from it.
A detection is only as good as the analysis. Findings from AI, thermal or LiDAR results inherit those analyses' limits; dismissing a false positive is part of the workflow.
Placement is an estimate with a stated radius. Without an RTK fix, or without a record of where the camera pointed, the radius is larger, and says so.
Grouping and matching are automatic and cautious. They may leave two findings for one object when the positions are uncertain; they never combine different kinds or reopen a closed finding.
A correction reaches a finding through a person's decision. A corrected detection changes a finding only when the person verifying chooses to keep it or to mark it a false positive; a finding nobody decided about is left unchanged.
Notifications are email only. Phone push notifications are not offered, and emails carry no images.
Offline on the phone covers the 25 most recently updated open findings assigned to you, and any finding opened before. Evidence images from encrypted files need a connection, and are never saved on the phone.
A certificate covers the history up to when it was signed. Changes after it need a new certificate.
Availability. The complete workflow, from every source through the mobile app, share links, webhooks, the API and certificates, is in production.
Where it matters
Utility inspection
Hot connectors from thermal, vegetation near lines from LiDAR and damage from AI, on one list, assigned to crews who close them out from the phone.
Solar operations
Hot modules and washouts found once, matched on every repeat flight, and tracked to repair with photographs from the field.
Roofing and insurance
Damage drawn on the photographs, placed on the roof, shared with the adjuster by link, and certified with its evidence.
Construction
Change alerts on earthworks become findings with the volume that changed, for the site team to confirm or dismiss.
Service providers
A pilot files findings in the job's folder; they arrive with the delivery, and the client works them in their own vault.
Claims and disputes
A finding whose every step is sealed, whose evidence is fingerprinted, and which a legal hold protects from deletion the day a dispute begins.
Standards
Standard
Used for
GeoJSON (RFC 7946)
Finding locations and areas, in exports and through the API
KML (OGC)
Export to Google Earth
CSV
Export to spreadsheets
HMAC-SHA256
Signing every webhook message, so your software can tell it came from Omniris
CAdES and PAdES (ES256), RFC 3161 timestamps
Signing and time-stamping finding certificates
OpenTDF
Encrypting evidence images cut from encrypted files
Glossary
Custody record
Omniris' tamper-evident log of who did what to a file or finding, and when; each entry is sealed so that a later change to it can be detected.
Evidence
What supports a finding: an image cut from a photograph or map, a whole file, a photograph from the field, with the values the source measured.
Finding
Something on a site that needs attention, recorded as a work item with a location, category, severity, status, assignee, evidence, comments and history.
Fingerprint
A short code computed from a file's exact contents; any change to the file gives a different fingerprint.
Grouping
Combining sightings of the same thing in one run into one finding.
Matching
Adding a new sighting to an open finding it falls on, instead of opening a second one.
Orthomosaic
Many survey photos stitched into one map-accurate aerial image.
Promote
Turn a detection you are looking at into a finding.
RTK
Real-time kinematic positioning: satellite positioning corrected to centimetre level.
Stage
One of four phases every finding is in: untriaged, active, resolved or dismissed.
Uncertainty radius
How far from its stated position a finding may really be, as a 95% estimate.
Verified
Said of an image whose AI detections a person has checked, corrected where needed and confirmed, with a sealed record of what changed.
Webhook
An automatic message Omniris sends to your own software's web address when something changes.
Omniris · Findings white paper, October 2026. This paper describes product capabilities. The example finding uses illustrative values.