Object detection built for vehicles, poles, assets, and reviewable counts.
Move from watching footage to working with measurable evidence.
Object detection makes large image and video sets easier to screen, compare, and act on. The strongest workflow keeps the visual proof connected to every count.
Faster first pass review
Surface target objects across long flights or wide sites so reviewers can focus attention where it matters.
Consistent visual labeling
Apply the same class definitions, mask colors, numbering rules, and confidence criteria throughout a dataset.
Countable outputs
Convert frame level detections into reviewed instance counts, persistent tracks, and site level summaries.
Traceable decisions
Keep annotated frames and video beside the count so clients can inspect why an object was included or excluded.
Repeatable comparisons
Apply a stable review method to repeat flights for progress monitoring, inventory changes, or site condition comparisons.
Structured handoff
Deliver object logs, timestamps, selected stills, and optional GIS ready records when source metadata supports coordinates.
Explore the renovation in 360°
Drag or swipe through this equirectangular interior panorama to inspect the exposed framing, ceiling structure, electrical runs, openings, and cyan instance segmentation evidence around the room.
See individual vehicles, not just pixels.
The example separates visible vehicle instances with cyan masks, maintains numeric IDs across motion, and displays a cumulative unique vehicle count for review.
- Parking inventory and utilization snapshots
- Traffic and access pattern observations
- Persistent IDs that reduce repeat counting
- Annotated evidence for stakeholder review
Build a faster, more reviewable first pass pole inventory.
The pole workflow highlights visible poles in cyan, assigns persistent numbers, and updates a running total as the aircraft moves through the corridor.
- Accelerate initial inventory formation across a route
- Keep each visible pole connected to a reviewable ID
- Support maintenance, attachment, and fieldwork planning
- Compare repeat flights for additions, removals, or visibility changes
- Prepare GIS ready records when geospatial metadata is available
Segment vehicles and buildings across an entire orthomosaic.
This model assisted, visually verified example separates 19 visible road vehicles and 22 fixed roof houses and buildings into individual masks. Stakeholders can trace each total back to the imagery and review how occlusion and excluded objects were handled before using the results for inventory, planning, or repeat site comparisons.
19
Visible road vehicles
22
Fixed roof buildings
Why stakeholders choose this service
- Trace every reported total to a visible instance mask
- Review vehicle activity and building inventory in one deliverable
- Apply clear class, exclusion, and occlusion rules across the site
- Prepare structured or GIS ready outputs when source geospatial metadata supports them
This sample excludes towable RVs and trailers, equipment, canopies, slabs, and uncertain fragments. Dense canopy is treated as occlusion; hidden pixels are not inferred.
Download the full resolution segmentation image (15.9 MB)
A target specific workflow for infrastructure, sites, and recurring visual review.
Each project begins with a precise object definition. That keeps the model, review rules, and deliverables aligned with the decision the client needs to make.
Vehicles
Parking counts, fleet observations, access studies, traffic snapshots, and visible utilization patterns.
Utility assets
Electric poles, selected attachments, corridor assets, and visually distinguishable utility inventory targets.
Construction activity
Equipment, selected material groups, staging areas, and repeat flight progress indicators.
Built assets
Roof elements, signs, site fixtures, structures, and other defined objects visible at the source resolution.
Inspection triage
Flag candidate objects or conditions for human review without presenting automated detections as final engineering findings.
Change monitoring
Apply the same class and review definitions to comparable capture sets for visible change analysis over time.
Four steps from source footage to reviewed object data.
Define the target
Agree on object classes, inclusion and exclusion rules, count logic, footage conditions, and the decision the output must support.
Test representative media
Evaluate resolution, altitude, angle, lighting, motion, occlusion, and class confusion on a representative sample.
Detect, segment, and track
Run the agreed workflow, attach IDs across frames where feasible, and generate the requested overlays and structured records.
Review and deliver
Inspect edge cases, document limitations, reconcile counts to the agreed QA level, and package the usable deliverables.
Choose the output your team can actually use.
Visible masks, boxes, class labels, persistent IDs, timestamps, and count overlays configured for the scope.
CSV or JSON records with IDs, classes, frame or time references, confidence fields, and review status where requested.
Project level totals with counting rules, known limitations, validation notes, and any unresolved manual review cases.
High resolution stills that show representative detections, edge cases, and agreed examples for stakeholder review.
Point or feature output when imagery includes sufficient coordinate, camera, calibration, and positioning metadata.
A concise record of target definitions, capture constraints, review level, and how automated results should be interpreted.
Common object detection questions
What can AeroFrohne detect in aerial imagery?
Common targets include vehicles, electric poles, site equipment, selected materials, structures, and other visually distinct assets. Feasibility depends on object size, image resolution, viewing angle, occlusion, lighting, and the available training or reference data.
How accurate are object detection and counting results?
Accuracy varies by footage and target class. AeroFrohne defines the target, tests representative footage, reviews edge cases, and documents limitations before a count is used for operational decisions.
Can the workflow count unique objects instead of counting every frame?
Yes. Persistent tracking IDs can reduce repeat counting as an object remains visible or reappears within the supported tracking window. Occlusion, camera cuts, long absences, and similar looking objects can still require human review.
What can be delivered for an electric pole count?
A typical package can include annotated MP4 video, a reviewed pole count, persistent IDs, selected still frames, and an object log. A GIS layer can be added when the source includes sufficient geospatial and camera metadata.
What footage produces the strongest results?
Stable, high resolution footage with consistent altitude, deliberate overlap, limited motion blur, and clear views of each target produces stronger results. Coordinates require calibrated or georeferenced source data rather than video alone.
Send the target, the media, and the decision your count needs to support.
AeroFrohne will review feasibility, define the object class and count logic, identify likely edge cases, and recommend the right output and QA level.