Dashcam Visual Analytics & Driver Scoring Platform

Frame-by-frame video analysis pipeline

Built an OpenCV-based ingestion and processing pipeline that walks vehicle dashcam footage frame-by-frame for stable, repeatable feature extraction before model inference.

Scene change detection, tracking, and event extraction

Combined scene change detection with object tracking to isolate meaningful segments and pull structured driving events from noisy real-world road footage.

PyTorch model fine-tuning for critical events

Fine-tuned PyTorch models for hard braking, lane departure, and tailgating detection, iterating on fleet-specific data to tighten precision and reduce false positives.

Driver behaviour scoring and timeline reports

Implemented behaviour scoring, per-trip timelines, and exportable reporting hooks via Django so trainers and compliance workflows could consume results without manual clip review.

Commercial fleets

Objective scoring from dashcam evidence for coaching conversations without all-day manual review

Road safety programs

Repeatable event detection for tailgating, hard braking, and lane departure across large video libraries

Compliance and training leads

Audit-friendly timelines and aggregates that align teams on the same safety metrics

UK operators

Region-specific deployment and analytics tuned to UK driving conditions and fleet policies