AI-assisted, human-confirmed
Detection proposes; inspectors decide. Every finding carries the reviewer, the evidence frame and the annotation that produced it.
NovaInspect
AI-Powered Inspection
From visual evidence to actionable reports.
Upload inspection video or imagery, let detection do the first pass, confirm findings by annotation, and generate the report — in one workflow, with the evidence attached at every step.
Inspection project — NH section review
Pothole · HighCrack · MediumAI detection · 2 findings · 1 pending review
| Finding | Severity | Chainage |
|---|---|---|
| Transverse crack | Medium | 12+450 |
| Pothole | High | 12+620 |
| Edge break | Low | 13+080 |
Visual inspection platform
Inspection evidence that holds up to review
NovaInspect is built around the way inspection work is actually signed off: a finding is only as good as the evidence and the reviewer behind it.
One structure for video and images
Frames pulled from video and standalone images enter the same project, so findings are comparable however the evidence was captured.
Report-ready by design
Findings, severity, location and evidence are recorded as data first, then composed into reports you can export and circulate.
Video to report
Turn inspection video into a structured report
Long survey runs become reviewable frames, detected findings and a report — without hand-scrubbing footage.
- Video upload
- Frame extraction
- Human annotation
- Report generation
Capture and prepare
Get long inspection runs into a reviewable form.
- Video upload per project or section
- Automatic frame extraction
- Frame selection and filtering
- Evidence frames retained with each finding
Detect and annotate
Model output and inspector judgement in the same pass.
- Object and defect detection on extracted frames
- Annotation on the exact frame under review
- Inspection comments against each observation
- Severity classification from your catalogue
Findings and reporting
From observation to a document someone can act on.
- Finding management and review status
- Report generation from confirmed findings
- Exportable reports for circulation
- Consistent structure across inspections
Image to report
Inspect image sets with the same rigour
For drone stills, handheld photos and site uploads, the path from image to finding to report is identical.
- Image upload
- AI analysis
- Annotation
- Report
Batch image inspection
Work through a set of images without losing context.
- Single and multiple image upload
- Inspection sets grouped by asset or location
- AI-assisted image analysis
- Evidence images attached to findings
Precise marking
Mark what you mean, at the geometry the defect needs.
- Bounding boxes for discrete defects
- Polygon annotations for irregular areas
- Point annotations for single locations
- Labels drawn from your defect list
Classification and output
Consistent labelling is what makes reporting comparable.
- Comments against each annotation
- Severity applied per finding
- Finding lists per asset or section
- Report generation and export
Annotations
Annotation is how inspection becomes data
Each mark carries geometry, a classification and a severity — which is what turns a reviewed image into a queryable, reportable record.
Bounding box
Rectangular extent around a discrete defect or object.
Polygon
Irregular area such as a patch, spall or delamination.
Polyline
Linear features — cracks, joints, cable runs, kerb lines.
Point
Single-location marks for assets, samples or observations.
Freehand
Traced outlines for shapes a polygon cannot follow cleanly.
Measurement
Lengths, widths and areas captured alongside the annotation.
Text / comment
Inspector notes attached to the exact evidence frame.
Defect classification
Label and severity applied from your own defect catalogue.
From a mark on an image to a structured finding
Annotations are stored against the asset and location they describe, so findings can be grouped, filtered, compared between inspections and exported without re-reading the imagery.
- Geometry retained with the annotation, not flattened into the image
- Classification and severity from your own catalogue
- Comments preserved for the review trail
- Findings grouped by asset, section or inspection run
Example finding record
- Asset / section
- NH-53 · Section 4
- Location
- Chainage 12+620
- Annotation
- Bounding box (frame 0142)
- Classification
- Pothole
- Severity
- High
- Evidence
- Extracted frame + source video reference
- Reviewer
- Confirmed by inspector
Illustrative record — fields are configured per inspection programme.
AI / ML model training
Models trained on your inspections, not someone else's
Annotation work feeds training, training produces versioned models, and corrected findings feed the next version.
Dataset
Collect the imagery and frames that represent the conditions you inspect.
Annotation
Label the dataset using the same annotation tools inspectors use daily.
Dataset validation
Check label consistency and coverage before anything is trained on it.
Model training
Train a model version against the validated training dataset.
Model evaluation
Score the version on a held-out set and record precision and recall.
Model version
Keep versions with their dataset, parameters and evaluation results.
Deployment
Promote a version so new inspections run inference against it.
Continuous improvement
Feed corrected findings back into the dataset for the next version.
Dataset management
Annotation datasets and training datasets are managed separately, so what a model learned from stays traceable.
Model versions
Each version records its training status, dataset and evaluation run — no guessing which model produced a finding.
Evaluation and inference
Precision and recall are reported per version against your own evaluation set, then inference runs on new inspections.
We do not publish blanket accuracy figures. Performance depends on the dataset, defect classes and capture conditions of each programme, and is reported per model version against your own evaluation set.
Road inspection
MoRTH-aligned road inspection workflows
Designed to support MoRTH-oriented road inspection and reporting practice, with chainage-based evidence for every defect raised.
- Road condition inspection from video or images
- Pavement defect identification and classification
- Crack detection and extent marking
- Pothole identification with evidence frames
- Road-surface assessment by section
- Road asset inspection alongside surface defects
- Chainage-based location for every finding
- Compliance-oriented inspection reporting
Project to report, by chainage
State road programmes are organised by project, section and chainage — for example a Chhattisgarh road project — so every finding resolves to a point on the network.
- Project
- Road section
- Chainage
- Captured evidence
- Detected defect
- Annotation
- Finding
- Report
Novametrics builds inspection workflows aligned to MoRTH-oriented reporting practice. Applicable standards, thresholds and acceptance criteria are configured from the specifications your authority issues — we do not substitute our own.
Ready to turn visual data into actionable intelligence?
Bring us a section of road, a building portfolio or an inspection backlog, and we will show you the NovaInspect workflow against your own evidence.
Talk to NovaMetrics