THE ISSUE, BROKEN DOWN
A Gap Between Current Capabilities and Operational Needs
Inspection Frequency
Current workflows rely on infrequent inspections, allowing defects to worsen unnoticed. CAG requires near-daily data collection for timely detection.
Integration & Scalability
Current solutions operate in silos with limited integration. A scalable, end-to-end system is needed to support fleet-wide monitoring and coordinated maintenance.
Detection Reliability
Manual inspections are inconsistent and prone to human error. More reliable, data-driven detection is needed to ensure accuracy across the network.
Maintenance Prioritization
Existing approaches lack clear severity assessment, making it difficult to prioritise repairs. CAG needs systems that translate defects into actionable maintenance tasks.
THE SOLUTION
CAPTURE
ANALYSE
MAINTAIN
A COST-EFFECTIVE SCALABLE SYSTEM FOR DAILY ROAD MONITORING
Detecting defects, quantifying severity, and delivering geo-tagged insights for proactive maintenance.
Enhanced Detection Reliability
Sensor fusion (vision + IMU + GPS) improves consistency beyond vision-only models (~70% baseline), with contextual severity validation.
Monthly → Daily Monitoring
Transition from periodic inspections to continuous, data-driven road condition tracking.
75+ km Airside Network Mapped
Scalable coverage across multiple zones, with capability to expand to additional road networks.
Cost Reduction vs Outsourced Inspections
Eliminates reliance on external vendors and reduces reporting delays, enabling faster and more cost-efficient maintenance cycles.
WHAT USERS SAY
-
“This solution could significantly shorten the defect detection-to-action cycle.”
-
“Geo-tagged evidence and severity levels would greatly improve maintenance planning.”
-
“Deploying this on operational vehicles is a practical approach for routine monitoring.”
-
“A defect dashboard would be highly valuable for tracking and recurring hotspot analysis, and is highly beneficial for both operations and long-term planning.”
-
“With strong accuracy and low false positives, this could become a valuable operational tool.”
SEE THE SYSTEM IN ACTION
EXPLORE THE TECHNOLGY
Deep dive into each component of our solution—from AI models to hardware sensors, dashboard analytics, and BIM integration.
Physical Sensor Mount
A precision-engineered PCB featuring GPS, IMU, and environmental sensors that capture high-frequency telemetry synchronized with video frames for accurate defect geotagging.
- 10Hz GPS with SBAS correction for sub-meter accuracy
- 6-axis IMU detecting vertical acceleration spikes
- Bluetooth Low Energy sync with mobile app
- Custom PCB with active cooling and UPS backup
Edge Capture App
Our custom Android application connects directly to the ESP32 hardware via Bluetooth to verify sensor status, initiate recordings, and sync journey data to the cloud.
Computer Vision & Spatial Mapping
Defect Detection: Identify 10 types of road and curb defects using a YOLOv12 model.
Precision Tracking: Track multiple defects across frames using BoT-SORT, with a short validation threshold to filter out false detections.
Precision Localization & Road Analytics
Geospatial Tagging: Fuse GPS and onboard sensor data (IMU) using an Extended Kalman Filter (EKF) to assign accurate coordinates to each detected defect, including improved positioning in low or no GPS environments (e.g., indoors).
Road Quality Insights: Uses IMU sensor data and visual analysis to evaluate road smoothness and detect rough segments.
Smart Event Detection: Automatically identifies bumps while filtering out normal vehicle movements like turns.
Smart Severity Control
Context-Aware Prioritization: Highlights more severe defects by considering overall road condition, so deteriorating areas get higher attention.
Sensor-Validated Alerts: Flags critical issues only when visual detection is reinforced by IMU sensor data.
Blind-Spot Coverage: Captures hidden or hard-to-see defects using IMU data, even when visibility is limited.
Smart Noise Filtering: Reduces false alerts by ignoring low-confidence detections unless supported by sensor data.
OPERATIONAL INTELLIGENCE DASHBOARD
A centralized operational platform that processes multi-modal sensor data (video, GPS, IMU) through a cloud-based ML pipeline to automatically detect, geolocate, and prioritise road defects, while enabling end-to-end workflow management from detection to repair.
The system includes fault-tolerant pipeline monitoring with continuous status tracking from data upload to database persistence, ensuring reliability and transparency across processing stages. It also supports on-site post-repair verification, allowing maintenance crews to upload before-and-after evidence, maintain a historical audit trail, and instantly update defect status upon completion.
Overview Page
Real-time visibility of network-wide defect status, highlighting urgent issues, active repairs, and critical segments through KPI summaries and an interactive map.
Alerts & Triage
Structured triage interface for reviewing new detections, validating defects, flagging false positives, and prioritising critical cases for action.
Defect Management
Advanced filtering and search across all defect records, with spatial visualisation, severity breakdowns, and direct assignment to maintenance teams.
Defect Analytics
Advanced analytics on defects with detailed visualisation such as distribution by types and zone location.
Roughness Analytics
Segment-level road condition analysis using fused sensor metrics, enabling identification of deteriorating areas and trend monitoring over time.
Coverage Monitoring
Inspection frequency tracking across the network, ensuring consistent monitoring and highlighting under-inspected segments.
Mobile-Optimised Maintenance Interface
To support maintenance technicians working on-site, the system provides a dedicated mobile-optimised Field Interface. Designed for quick access to assigned repair tasks, location-based navigation, and image-based maintenance updates while in the field. It allows for direct upload of before- and after-repair images using phone’s camera as required evidence before status transitions.
BIM DEFECT VISUALIZATION
Seamlessly export defect data into Autodesk Revit as native BIM families, allowing engineers to visualize infrastructure health directly on the airport’s 3D digital twin.
• Defects rendered as 3D markers with metadata
• Overlay historical data for long-term planning
• one-click extension installation (via .exe)
SYSTEM ARCHITECTURE
A seamless pipeline from the physical edge to the digital dashboard, engineered as a distributed system.
USER JOURNEY
A seamless pipeline from the physical edge to the digital dashboard, engineered as a distributed system.
Operator drives the airfield with our mobile app and sensor suite. Video and IMU/GPS data are captured automatically.
Data is uploaded to AWS cloud storage. Sensor telemetry is synchronized with video frames using timestamps.
Our computer vision model processes video to detect defects. GPS/IMU fusion pinpoints exact geolocation and severity.
Detected defects appear on the dashboard with location, severity, and images. Managers receive real-time notifications.
Dashboard aggregates defects, calculates roughness index, and helps managers assign repair tasks by priority.
Maintenance teams receive work orders. Check assigned defects via dashboard and upload before and after pictures.
Historical defect data is exported to Autodesk Revit for long-term infrastructure planning and 3D visualization.
What used to take weeks of manual reporting now happens in hours. Operators drive their usual routes, and managers get instant, actionable insights—no clipboards, no guesswork.
PROJECT ROADMAP
Discover: Week 1–2 (Term 7)
Conducted stakeholder interviews and field observations to uncover key pain points in current road inspection workflows, including delayed reporting, subjective defect classification, and lack of maintenance visibility. In parallel, explored the solution space by evaluating different technology stacks—such as computer vision models, IMU-based sensing, GNSS localization, and cloud processing pipelines—to identify feasible and cost-effective approaches for automated defect detection.
Define: Week 3–4 (Term 7)
Synthesized research insights into a clear problem statement and defined system requirements, focusing on improving detection accuracy, severity quantification, and actionable maintenance insights.
Development Iteration 1 (Prototype Foundation): Week 5–7 (Term 7)
Built initial proof-of-concept with smartphone-based data collection (camera + IMU + GNSS), early YOLO model training, and a basic dashboard for defect tracking and visualization.
Development Iteration 2 (System Integration): Week 1–3 (Term 8 )
Improved system reliability through cloud integration, wireless data upload, UI redesign, and hardware refinements including power management and thermal control.
Development Iteration 3 (Model & Data Refinement): Week 4–6 (Term 8)
Enhanced computer vision accuracy and IMU-based roughness detection through better data labeling, model tuning, and improved signal processing for more reliable defect detection.
Development Iteration 4 (Validation): Week 7–9 (Term 8)
Focused on system validation and testing end-to-end performance under real operational conditions.
Final Solution (Integrated System): Week 10–12 (Term 8)
Delivered a fully integrated, cost-effective road inspection system combining computer vision and sensor fusion, with a centralized dashboard enabling real-time monitoring, severity assessment, and proactive maintenance planning.
Building Smart & Sustainable
Our approach prioritizes resource efficiency, cost reduction, and environmental responsibility.
Resource & Cost Efficiency
IRIS reduces the environmental and financial footprint of road maintenance by automating inspections and eliminating paper-based workflows
- 60% reduction in inspection vehicle mileage
- Eliminated ~10,000 sheets of paper per year
- Lower fuel consumption and carbon emissions
Synthetic Data Generation
To reduce the need for expensive real-world data collection, IRIS uses generated synthetic airside road imagery using Unreal Engine and procedural defect modeling.
- Trained model on 8,000+ synthetic images before real data
- Reduced real-world data requirements by 40%
- Faster iteration cycles during development
Iterative Fabrication & FEA
IRIS’ Sensor Suite went through 4 design iterations using rapid prototyping and Finite Element Analysis to optimize strength-to-weight ratio and thermal performance.
- 3D-printed enclosures with PETG for durability
- FEA simulations prevented overheating failures
- Reduced material waste by 55% vs. traditional prototyping
