Computer vision data pipeline: ingest street video, run YOLO detection, and publish reliable crossing metrics through an AWS + Airflow workflow.
YOLO detection and crossing event tracking
Production-style pipeline that converts raw video files into structured crossing metrics.
Data engineering work included ingestion, ETL, workflow orchestration, cloud storage, and analysis-ready exports.
Source code is private due to internship data and cloud environment constraints.
Collect video files and register metadata for tracked processing.
Prepare frames with consistent format and quality checks.
Detect people and capture movement events per zone.
Aggregate detections into crossing counts and timestamps.
Run scheduled DAGs with retries, logs, and rerun control.
Publish parquet/CSV outputs for SQL analysis and dashboards.