Smart Pothole Detection automatically identifies and confirms road defects using crowdsourced smartphone sensors, AI-powered spatial analysis, and a multi-signal ML confidence engine — deployed entirely on the cloud.
Laser profilometers and LiDAR survey vehicles provide high accuracy but cost tens of thousands per survey run — covering only a fraction of road networks.
Simple threshold-based smartphone apps flag every speed bump, rail crossing, and rough patch as a pothole — producing noisy, unreliable datasets unusable for planning.
Citizen report platforms like FixMyStreet rely on subjective human reports, creating geographic bias toward digitally active areas and leaving rural roads unmapped.
Deep learning approaches require large annotated datasets of GPS-tagged, field-verified pothole locations — a substantial collection effort that limits practical deployment.
The Smart Pothole Detection mobile app samples accelerometer and gyroscope at 10 Hz and GPS at 1 Hz. Readings are batched and sent to the cloud every 3 seconds from any number of participating devices.
A cloud ingestion service validates every reading and computes AI-derived features from the raw sensor data before streaming to our data lake — enriching each reading for downstream ML analysis.
Every 30 minutes, a batch ML job applies AI-powered spatial clustering to GPS coordinates, then scores each cluster using a multi-signal confidence engine — automatically distinguishing genuine potholes from speed bumps and road noise.
Confirmed potholes are served via a REST API to an interactive Leaflet dashboard, with severity-coded markers, proximity search, and status tracking. Weekly PDF and CSV reports are emailed automatically to stakeholders.
Raw sensor readings vary significantly depending on vehicle speed and device orientation. Our AI normalises and enriches every reading in real time — making sensor data from any driver, any speed, any device directly comparable for accurate analysis.
Our ML engine groups GPS readings by geographic proximity, intelligently adapting to the density of any road network — whether a busy city junction or a quiet rural lane. Both are detected accurately without manual configuration.
AI and ML are used to distinguish genuine potholes from speed bumps, rail crossings, and road noise — with no labelled training data required. The system learns the difference from the physical characteristics of each impact pattern.
Every pothole candidate is evaluated by multiple independent AI signals. A composite ML confidence score determines whether a candidate is confirmed — filtering out false positives while ensuring genuine defects are never missed.
iOS and Android app collects sensor data passively. Any device can contribute — the more drivers, the better the coverage.
Leaflet map with severity-coded markers, cluster overlays, proximity search, and real-time auto-refresh every 60 seconds.
Authenticated API with GeoJSON export, spatial queries, and status lifecycle management. Integrate with any GIS or asset system.
Weekly PDF and CSV reports emailed automatically to stakeholders every Monday. On-demand generation via API.
AI-powered spatial matching merges multiple reports of the same defect into a single, progressively enriched record — eliminating duplicates automatically.
Physics-informed signal engineering means the system deploys immediately in any city with zero historical training data.
Serverless Cloud Run, BigQuery streaming, Cloud Scheduler, and Terraform IaC. Scales from a single-city pilot to a national deployment.
Confirmed potholes are automatically classified as HIGH, MEDIUM, or LOW severity based on peak impact readings, enabling prioritised maintenance scheduling.
Replace expensive survey vehicles and citizen hotlines with a continuous, AI-driven monitoring pipeline. Receive weekly reports and real-time alerts for newly detected high-severity defects. Prioritise repair schedules by severity and report count.
Monitor large road networks continuously without deploying specialised vehicles. AI and ML are used to handle everything from busy motorways to quiet rural B-roads automatically, scaling to any network size.
Access an evidence base of road defect locations, severities, and persistence data to support road safety campaigns, insurance claims, and advocacy for infrastructure investment in high-risk areas.
Serverless containers. Auto-scales from 0 to 10 instances. Zero cold-start cost when idle.
Day-partitioned sensor data lake. Sub-second streaming inserts. Scales to billions of readings.
Spatial database for confirmed potholes. ST_DWithin deduplication. Private VPC access.
All credentials stored in GCP Secret Manager. Injected at runtime. Never in environment files.
Entire infrastructure defined in code. Reproducible deployments from pilot to national scale.
ML job triggered every 30 minutes. Weekly report delivery every Monday at 08:00 UTC.
Xenova Systems Limited is a technology company specialising in AI-powered infrastructure monitoring solutions. Smart Pothole Detection is our flagship product — a cloud-native platform that brings machine learning and crowdsourced sensing to one of the most persistent infrastructure challenges facing local authorities worldwide.
Our approach is grounded in physics-informed signal engineering: rather than relying on expensive labelled datasets or black-box deep learning, we build systems where every algorithm decision is explainable, every threshold is calibrated to physical reality, and every deployment is reproducible through infrastructure-as-code.
A Cloud-Native Crowdsourced System Using OPTICS Clustering and Multi-Signal Confidence Scoring
Read the Paper →Get in touch with the Xenova Systems team to discuss a pilot deployment, API access, or a custom integration for your road network.