01 · Challenge 02 · Solution 03 · Demo 04 · Architecture 05 · Models 06 · Results 07 · FAQ
01 / The Challenge

Why is highway safety auditing so difficult to scale?

India operates one of the largest road networks in the world, and every kilometre of it depends on a precise inventory of road safety features — crash barriers, signs, lane markings, delineators, lighting, and pavement condition — to keep drivers safe and operators compliant with IRC and MoRTH standards.

Auditing those features today is overwhelmingly manual: survey teams drive routes with clipboards and cameras, log conditions by hand, and produce static reports days or weeks after the inspection. With over 150,000 road fatalities annually and rising regulatory expectations, manual highway audits have hit a scalability ceiling.

Highway authorities, municipal corporations, construction firms, and fleet operators all face the same set of structural challenges. AI-driven computer vision finally turns one-off audits into a continuous, scalable, evidence-based capability — and Drishti is what that looks like deployed.

Costly manual inspection

Survey teams driving long routes with clipboards and cameras generate high labor costs per kilometre audited.

Slow audit-to-report cycle

The lag between inspection and report leaves authorities and operators acting on stale data.

Inconsistent results

Subjective scoring and varying field practices produce audits that aren't directly comparable across regions or auditors.

Delayed defect detection

Critical hazards go unnoticed between scheduled inspections, increasing accident risk.

Limited scalability

Manual processes simply cannot match the growth of national, state, and concessionaire road networks.

Mixed data quality

Field footage varies in resolution, lighting, weather, and camera quality — naive ML pipelines fail under that variation.

02 / The Solution

How does Drishti's road safety AI work?

Drishti is an AI-driven computer vision platform that turns standard vehicle dashcam or smartphone video into a continuous stream of structured highway intelligence. The system detects, segments, and classifies road safety features along any highway, audits them against IRC and MoRTH standards, and produces evidence-based recommendations — all at real-time inference speeds of 45–60 FPS.

It covers 8+ asset categories from a single video pass: pothole detection and segmentation (85%), road-surface classification across concrete, asphalt, gravel and damaged (97%), five-class crack classification (82%), crash barrier and signage monitoring, lane and edge-line detection, delineators, raised pavement markers, lighting and signals, kerb and shoulder assessment (71%), and pavement and drainage condition flags.

An ensemble of YOLOv8-L, YOLOv11-L, and a lightweight YOLOv11n classifier — with routing that sends simple lookups to small models and hard frames to large ones — runs at 45–60 FPS on standard cameras. Every detection carries GPS coordinates and a confidence score, feeding an audit dashboard with KPIs, trend lines, and compliance reports formatted for NHAI, state authorities, and concessionaires.

YOLOv8-L YOLOv11-L YOLOv11n Computer Vision Deep Learning Python AWS PostgreSQL MongoDB
03 / Project Demo

See Drishti in action

A walkthrough of the Drishti road safety audit — from dashcam or smartphone video ingestion through the YOLO computer-vision ensemble to the auditor dashboard surfacing GPS-tagged defects, confidence scores, and surface classification in real time.

Live detection overlay on road video
Pothole, crack & surface classification
GPS geo-tagged defect logging
Severity & affected-area metrics
04 / Architecture

What is the architecture of Drishti's road safety platform?

A five-stage pipeline — from highway data capture, through cloud ingestion, AI computer vision, and audit & compliance intelligence, to stakeholder-facing applications. Cloud-native, horizontally scalable, and validated for Indian highway conditions.

01 ▸ CAPTURE
Highway Data Capture
Standard dashcam or smartphone video across road types, regions, and conditions — frame-sampled with per-frame GPS.
02 ▸ INGEST
Cloud Ingestion
Cloud-native pipeline with secure storage and a unified analytics layer; horizontal scaling for multi-route surveys.
03 ▸ VISION
AI Computer Vision
YOLOv8-L + YOLOv11-L detection and a YOLOv11n classifier detect, segment, and classify 8+ asset categories.
04 ▸ AUDIT
Audit & Compliance Intelligence
Detections audited against IRC, MoRTH, and NHAI standards, with confidence scoring and evidence-based recommendations.
05 ▸ DELIVER
Stakeholder Applications
Auditor dashboard, geo-map view, KPIs and trend lines, plus compliance reports for NHAI, state authorities, and concessionaires.
Drishti Road Safety Audit — Computer-Vision Defect Detection Architecture YOLOv8-L + YOLOv11-L detection & segmentation with a YOLOv11n surface classifier over standard vehicle imagery, geo-tagged into audit-grade reports 1. HIGHWAY CAPTURE Dashcam / Smartphone Video • Dashcam / survey video • Frame sampling 45–60 FPS • GPS coordinates • IST timestamps • Varied lighting & weather • Indian road textures Coverage: standard cameras, no specialised survey hardware required 2. CLOUD INGESTION Cloud-Native Pipeline • Frame extraction • Image augmentation • S3 image storage • PostgreSQL + MongoDB • EC2 GPU compute • REST API endpoints Turns raw video into a clean, geo-tagged frame set ready for inference 3. AI COMPUTER VISION YOLOv8-L + v11-L + v11n • YOLOv8-L detection • YOLOv11-L segmentation • YOLOv11n classifier • Instance masks • 8+ asset types • 34,540+ training images Three YOLO models ensemble to detect, segment and classify defects in real time 4. AUDIT & COMPLIANCE IRC · MoRTH · NHAI • IRC / MoRTH audit checks • NHAI compliance scoring • Severity & confidence scoring • Geo-tagged defect map • Affected-area metrics • Safety-asset gap flags Converts detections into prioritised, evidence-based maintenance actions 5. STAKEHOLDER APPS Dashboards & Reports • Interactive Folium maps • Condition dashboards • Priority ranking • Automated reports • Compliance documentation • Pre / post-repair views Consumers: highway authorities, municipal, construction and fleet operators DELIVERY PHASES 1. Data foundation & labeling 2. IRC / MoRTH rule library 3. Transfer-learning training 4. Held-out highway validation 5. Edge & dashboard rollout 6. Pilots → Audit-as-a-Service Codified IRC, MoRTH, and NHAI provisions into the audit rule library; validated on held-out highways and piloted on priority corridors before broader rollout PLATFORM CAPABILITY PILLARS MIXED DATA QUALITY • Robust mixed-quality layer • Multi-frame fusion • Confidence scoring • Augmentation-heavy training • Mixed-resolution handling • Lighting, weather & night robust • Dashcam & smartphone video • Generalizes across cameras • No special hardware CV MODEL ENSEMBLE • 85% pothole detection • 97% surface classification • YOLOv8-L detection • YOLOv11-L segmentation • YOLOv11n classifier • 5-class crack typing • Instance segmentation masks • Real-time 45–60 FPS • 8+ infrastructure types SCALE & COMPLIANCE • Cloud-native horizontal scaling • Edge inference on vehicles • Async, parallel pipelines • API-first integration • IRC / MoRTH / NHAI audit logic • Standard cameras, no survey fleet • Phased route-by-route rollout • Open-source backbones • Compliance reports for NHAI MEASURABLE OUTCOMES • 85% pothole accuracy • 97% surface accuracy • 8+ asset types audited • Targeted, not blanket, repairs • Continuous network coverage • Geo-tagged audit trail • Compliance documentation • Proactive hazard detection • Built by AiSPRY Drishti · YOLOv8-L + YOLOv11-L + YOLOv11n · Computer Vision + Python · AWS + PostgreSQL + MongoDB · Built by AiSPRY

Five-stage pipeline · Six delivery phases · Four capability pillars

05 / Model Families

What models power Drishti?

Drishti routes the workload across an ensemble — small models for simple lookups, large models for hard frames — trained on 34,540+ images (27,690 train · 3,330 validation · 3,520 test) for accuracy across 8+ asset categories at real-time speed.

High-Accuracy Detection

YOLOv8-L

High-accuracy object detection that locates potholes, cracks, signage, and safety assets in each frame — the workhorse that finds where the defects are.

Next-Generation Detection

YOLOv11-L

Next-generation detection performance with segmentation for precise boundaries, so affected area and severity can be quantified — not just a box, but the shape of the damage.

Surface Classifier

YOLOv11n

A lightweight classifier for road-surface classification at scale — concrete, asphalt, gravel, damaged — at 97% accuracy with minimal compute, fast enough to run alongside the detectors.

06 / Results

What measurable results does Drishti deliver?

Drishti was designed to move three things at once — highway safety compliance, manual inspection cost, and audit-driven revenue — in the same direction, validated on Indian highway conditions.

Highway safety & compliance

  • Detects road safety features on any highway in India
  • 85%+ pothole detection and 97%+ surface classification accuracy
  • Faster, more timely detection of safety gaps and hazards
  • Consistent, comparable audit outputs across regions and auditors
  • Audit logic aligned with IRC, MoRTH, and NHAI standards
  • Confidence score on every detection tells auditors what to verify

Cost & operational efficiency

  • Reduced manual inspection cost across the surveyed network
  • Lower labor effort per kilometre audited
  • Faster audit-to-report turnaround with automated reports
  • Scalable across road networks without linear headcount growth
  • Reuses standard cameras — no specialized survey vehicles
  • Phased, route-by-route rollout aligns spend with proven ROI

Coverage & reach

  • 8+ asset categories detected from a single video pass
  • 45–60 FPS real-time inference on standard cameras
  • GPS-tagged detections enable per-kilometre asset inventories
  • Cloud-native deployment supports fleet-wide rollouts
  • Edge inference on dashcams and smartphones for the field
  • Compliance reports for NHAI, state authorities, and concessionaires
07 / Frequently Asked

Questions about the system

What is Drishti?
Drishti is an AI-powered road safety audit and infrastructure intelligence platform built by AiSPRY using computer vision and deep learning. It detects, segments, and classifies road defects and safety infrastructure across 8+ asset categories from standard vehicle dashcam or smartphone video, at 45–60 FPS in real time.
What technology powers Drishti?
Drishti is built on an ensemble of YOLOv8-L for high-accuracy object detection, YOLOv11-L for next-generation detection, and a lightweight YOLOv11n classifier for surface classification. The models are trained on 34,540+ images (27,690 train, 3,330 validation, 3,520 test) and deployed for real-time inference at 45–60 FPS on standard cameras.
What can Drishti detect?
Drishti detects 8+ road infrastructure asset categories: potholes (85% accuracy), road surface classification (97% accuracy), five-class crack classification (82% accuracy), kerb assessment (71% accuracy), crash barriers, road signs, lane and edge markings, delineators and reflective markers, lighting and signals, and pavement and drainage condition.
How accurate is Drishti?
Drishti achieves 85%+ accuracy on pothole detection, 97%+ on surface classification, 82%+ on crack typing, and 71%+ on kerb assessment. Every detection comes with a confidence score so auditors know which to verify, and the system is validated against held-out Indian highway data.
Does Drishti need specialized hardware?
No. Drishti is designed to operate on standard vehicle dashcams and smartphone cameras. There is no requirement for specialized survey vehicles or LIDAR-grade hardware — though optional LIDAR depth signals can be ingested when available.
— Continuous, AI-driven road intelligence

Replace manual highway audits with continuous road intelligence.

Talk to the Drishti team to learn how computer vision can transform road safety auditing in your region or network.

Get Started Today Read the Case Study →