Traffic Object Tracking

Real-time vehicle & pedestrian tracking with live traffic congestion classification.

What it does

Tracks vehicles and pedestrians in traffic video (each counted once, not once per frame), classifies congestion as LOW / MODERATE / HIGH, and renders an annotated video plus a Streamlit dashboard over the results.

Quickstart

python -m venv .venv && source .venv/bin/activate   # .venv\Scripts\activate on Windows
pip install -e ".[dev]"

python -m traffic_intelligence run --input data/raw/avenue.mp4
python -m traffic_intelligence dashboard

CLI

Subcommand	What it does
run	Full pipeline over a video: tracking, counting, congestion level, annotated video, CSV/JSON exports.
analyze	Recomputes counts, tracked-activity span and speed statistics from an exported tracks.csv, without re-running the model.
dashboard	Launches the Streamlit dashboard over the outputs.

python -m traffic_intelligence analyze --input outputs/tracks/tracks.csv

Configuration

configs/default.yaml ships with general-purpose defaults (model size, detection resolution, congestion thresholds) meant to run reasonably on typical hardware, CPU included. Traffic density in particular is scene-dependent — a 6-lane avenue and a quiet side street don't hit "busy" at the same vehicle count — so after your first run, check outputs/analytics/tracks.csv (or summary.json) and adjust congestion.density_thresholds in your own copy of the config (--config path/to/yours.yaml) so LOW/MODERATE/HIGH line up with what your footage actually shows. Each setting in the file is commented with what it trades off.

Stack

Python 3.11+ · Ultralytics YOLO (ByteTrack/BoT-SORT) · OpenCV · Pydantic · Pandas · Streamlit/Altair · pytest

Status

This project is actively being developed and improved. If you're interested in contributing, see CONTRIBUTING.md or reach out at felibridge49@gmail.com.

License

MIT