Smart Parking Access Control (SPAC)

license plate detection, OCR extraction, resident matching, and access decisions, all from a single CLI.

final report

The final report can be found in docs/report/report.pdf

setup

python -m pip install -r requirements.txt

for kaggle downloads:

python -m pip install kaggle

quick start

python -m src.main --help

the cli reads defaults from configs/default.yaml. pass action flags to run one or more stages:

python -m src.main --infer                          # run inference on default test split
python -m src.main --infer --eval --report          # infer + evaluate + report
python -m src.main --all                            # full pipeline end to end

actions

flag	stage	module
--download	download + unzip kaggle dataset	kaggle cli
--seed-db	generate sample resident db	src.data.create_sample_db
--prepare	convert raw data to yolo format	src.data.prepare_kaggle_car_plate_dataset
--train	train plate detector	src.train.train_detector
--infer	run end-to-end inference	src.infer.run_inference
--eval	evaluate predictions	src.eval.evaluate_pipeline
--report	generate markdown tables	src.eval.generate_report_tables
--all	run all stages in order	-

input overrides

python -m src.main --infer --image path/to/image.jpg            # single image
python -m src.main --infer --image-dir path/to/images           # directory
python -m src.main --eval --use-ground-truth                    # with ground truth
python -m src.main --eval --ground-truth-csv path/to/gt.csv     # custom gt path
python -m src.main --config other.yaml --infer                  # custom config

outputs

action	output
--prepare	data/processed/car_plate_kaggle/ (yolo splits + dataset.yaml)
--train	runs/detect/outputs/detector_train/ (weights, curves, results.csv)
--infer	outputs/predictions/inference_results.json, outputs/demo/
--eval	outputs/metrics/evaluation_summary.json, evaluation_rows.csv
--report	docs/evaluation_tables.md

project structure

src/
  main.py            # cli entrypoint
  data/              # dataset preparation & resident db
  train/             # detector training
  infer/             # inference pipeline & preprocessing export
  eval/              # evaluation & reporting
  ocr/               # multi-backend ocr engine
  matching/          # resident plate matcher
  utils/             # shared helpers
configs/
  default.yaml       # central config for all stages

each module under src/ has its own README with detailed usage.

config

all default paths and thresholds live in configs/default.yaml. prefer editing config over passing long CLI arguments. key sections: paths, dataset, training, inference, ocr, evaluation.

git and raw images

data/raw/* and data/processed/* are gitignored, please download datasets locally.

Notes

• if models/plate_detector.pt is missing, inference will fail with a clear error.
• if OCR confidence is low, matching may return denied as expected.
• fuzzy matching can be toggled in configs/default.yaml.

Authors

Arad Fadaei, Sia Tedy