Drone Inspection Software: Features That Matter After the Flight

On This Page

• Understand what happens to the data
• Specify deliverables before comparing features
• Verify measurement, thermal, and AI claims
• Make the maintenance connection concrete
• Count costs through retention and exit
• Run a demonstration that ends outside the platform

Understand what happens to the data

There are three jobs to distinguish, even when one platform offers all of them.

Processing converts captures into usable outputs. Photogrammetry reconstructs geometry from overlapping photographs; its outputs can include a point cloud, a collection of 3D points, or an orthomosaic, a georeferenced image map. A textured mesh connects reconstructed geometry into surfaces for viewing. These outputs have different uses and should remain distinguishable from the original photographs.

Inspection review lets a person locate relevant evidence, annotate an observation, assign a condition category, and retain the reasoning behind a finding. For example, Flyability describes Inspector Desktop tools for reviewing footage, working with points of interest and annotations, and producing reports. Its online offering adds capabilities including comparison between inspections. Those are provider-described functions, not proof of performance on your assets. See the Inspector feature description for the product boundary.

Asset workflow connects reviewed findings to the owner's asset register and follow-up work. A flight folder is useful to the pilot; a maintenance team needs the component identity, location, current condition, evidence, and action owner. Ask which system owns each field and which one is the authoritative record after handover.

For occasional visual inspections, organized originals, a structured finding register, and a readable report may be sufficient. Repeat inspections across many assets make persistent component IDs, history, and integrations more valuable. Choose the required workflow first; do not make a 3D reconstruction mandatory if it adds no useful location or measurement information.

Specify deliverables before comparing features

Ask for both a readable report and data that can be used elsewhere. The report explains the finding; the export preserves the records and files needed to revisit it.

Esri's Site Scan export documentation distinguishes a PNG orthomosaic preview for presentation from the TIFF used for raster analysis. It also describes LAS and compressed LAZ point-cloud exports, with a reminder to check the receiving software's LAZ support. A format name on a feature list is only the start of the compatibility check.

Use this matrix to specify the handover. The file examples draw on Esri's export documentation; the checks are editorial recommendations to adapt to your asset and receiving systems.

Check export completeness explicitly. In Site Scan's measurement export instructions, CSV is limited to points, and only measurements enabled on the map are included in the export. This is a concrete example of why “CSV export” does not necessarily mean every object and field will leave the platform.

Ask for record counts before and after export, plus a list of unsupported fields. If severity, comments, or image references need a separate export, include that step in the handover procedure and cost estimate.

Verify measurement, thermal, and AI claims

A detailed model does not establish measurement accuracy

If software will support dimensions or change measurements, specify the quantity, permitted error, units, and relevant reference system before judging the demonstration. Ground sample distance describes the ground represented by an image pixel. It does not independently establish positional accuracy.

The USGS imagery-calibration report explains that ground control and tie-point quality contribute to geometric accuracy regardless of ground sample distance. It also recommends retaining calibration information and uncertainty in metadata. Ask for independent check results appropriate to your measurement task, and distinguish those from how closely the processing fitted the control data used to build the model.

For repeat inspections, require the reviewer to account for alignment, capture geometry, coverage, and processing differences before calling an apparent change physical deterioration. Preserve the earlier processing version. Reprocessing an old dataset should not silently replace the baseline used in an earlier decision.

The six levels of drone-inspection evidence can help your team separate a visible observation from a measurement or diagnosis before specifying software outputs.

Thermal support must preserve the required data

Radiometric files retain data used for temperature analysis. A colored thermal image can provide a useful visual overview while lacking the values needed for quantitative work.

Pix4D's current PIX4Dmapper thermal-processing guidance distinguishes legacy supported radiometric formats from newer formats that it cannot process radiometrically. It also explains that thermal JPEGs processed as ordinary RGB images produce visual maps without extractable temperature values. Support differs across the company's applications and versions, so a general promise of “Pix4D compatibility” is insufficient.

Request a trial with files from your exact sensor and capture settings in the proposed software version. Confirm whether the result contains temperature values, which units and corrections apply, and whether those values survive export. Have the thermography specialist determine the settings and operating conditions needed for interpretation. A successful import alone does not answer those questions.

AI results need a defined task and a review route

Detecting a component, classifying a defect, and flagging an unusual image are different tasks. The InsPLAD power-line inspection study evaluates those tasks separately and describes challenges including occlusion, varied lighting, perspective, and objects appearing at different scales. Its results concern its own dataset; they do not validate a commercial system on another asset class.

When a supplier presents an accuracy claim, ask what was counted, which defect classes were included, and how the evaluation data were separated from training data. Request the numbers of missed defects and false alarms by class, together with the conditions represented. One aggregate percentage can conceal the failure mode that matters most to your operation.

Keep suggested findings distinct from reviewer-accepted findings. The demonstration should show how a reviewer rejects an incorrect suggestion, records a missed condition manually, and retains the original evidence. Treat images with inadequate visibility as not assessed, rather than automatically healthy.

Make the maintenance connection concrete

A geographic information system, or GIS, organizes spatial information. A computerized maintenance management system, or CMMS, organizes maintenance work. An enterprise asset management system, or EAM, may hold the wider asset record. Agree which system receives the observation and which one controls the resulting action.

For a hypothetical roof finding, the handoff might carry an asset ID, roof-section ID, finding ID, observation text, evidence references, reviewer status, and follow-up owner. Those are proposed fields, not a report of an actual defect. Have your receiving-system administrator map them before accepting an integration claim.

For an application programming interface, or API, request the exact supported operations. Can it transfer attachments as well as text? Create and update a finding? Retrieve review history? Return the maintenance work-order ID? Determine whether the connector is supplied, configured for your system, or still requires development. Check transfer limits, version support, and whether evidence links expire before the maintenance record does.

Test three failure cases: resend the same finding, interrupt a transfer, and update a reviewed record. The required behavior is no duplicate work order, a recoverable failed transfer, and a clear rule for which system's revision wins. Specify how errors become visible and who resolves them.

Access controls also belong in the demonstration. Use representative administrator, reviewer, and external-client roles to verify viewing, editing, downloading, and sharing permissions. Obtain written answers on hosting location, retention, backups, restoration, subcontractor access, and use of your data for model training. A general security statement does not answer those operational questions.

Count costs through retention and exit

Compare proposals over the same intended term and workload. Ask each supplier to identify the charging unit: users, assets, projects, processing allowance, storage, automated analysis, or a combination. Establish how reviewers and client viewers are counted, and which export or integration functions require an additional entitlement.

A useful planning formula is:

Total software ownership cost over the chosen term = setup and migration + subscription and usage charges + review and data-preparation labor + integration and support + retention, export, and exit costs.

This is a budgeting structure, not a quoted price or market average. Populate it with written supplier terms and your own measured staff effort. Include conversion, manual relabeling, exception handling, retraining, and the infrastructure needed for any desktop processing. Do not count the same labor under both setup and recurring work.

Measure elapsed delivery time separately from staff time. Processing may run unattended, while a short export may still require repeated manual corrections. Time a representative job through accepted handover and record which stages consume staff effort. That gives a better basis for evaluating a savings claim than processing speed alone.

Before signing, settle what happens at cancellation: how long export remains available, whether old reports remain accessible, what format the archive uses, and whether bulk download or assistance costs extra. Include a rehearsal of that archive retrieval in the evaluation. Data ownership language is useful only alongside a workable way to obtain and use the data.

Run a demonstration that ends outside the platform

Give shortlisted suppliers the same representative captures, asset register, required fields, and downstream import instructions. Include difficult imagery and an earlier visit if repeat inspection is part of the job. Have the asset specialist define what the sample supports before using it to judge automated results.

Then follow one finding through the complete sequence:

1. Import the captures and identify any rejected files or lost metadata.
2. Locate the component and link the finding to its original evidence.
3. Review the observation, revise it, and retrieve its decision history.
4. Export the report, structured record, attachments, and required spatial or thermal data.
5. Import the finding into the receiving system and exercise the duplicate and failed-transfer cases.
6. Retrieve the same finding from the exit archive and, where required, compare it with the previous visit.

Choose according to the work that survives this demonstration. A small visual-inspection program may need a dependable register and file handoff. A repeated measurement program needs demonstrated quality controls and baseline management. A multi-asset maintenance program needs stable identities, review history, and dependable integration. If the proposed system cannot preserve the required evidence or complete the receiving-system handoff, resolve that gap before committing to it.

Source notes

• Flyability: Inspector. Provider feature description for desktop and online inspection review; not independent performance evidence.
• Esri: Export ortho, point cloud, and mesh. Technical documentation distinguishing presentation previews, analytical rasters, and 3D exports.
• Esri: Export measurements. Technical documentation of export formats and measurement-selection limits.
• USGS: Guidelines for Calibration of Uncrewed Aircraft Systems Imagery. Government technical report on calibration, geometric accuracy, and supporting metadata.
• Pix4D: Processing thermal images, PIX4Dmapper. Product documentation of radiometric support and visual-only thermal outputs.
• Vieira e Silva and colleagues: InsPLAD. Research manuscript describing separate computer-vision tasks and dataset limitations for power-line inspection.

Last checked: September 7, 2026.