Pixels
Binary Image
Sample Parking Lot 22
User-End

ParkAssist — UMBC Parking Occupancy MVP

Author

Author: HackUMBC 2025 - Team GoGreen - Khushika Shah (kshah7@umbc.edu) - Rajjya Rohan Paudyal (rajjyap1@umbc.edu)

One-liner: ParkAssist gives UMBC students a live, privacy-safe count of open parking spaces—shown as Lot 22(A): free/total—so they stop circling and get to class on time.

Overview

Classic OpenCV pipeline (no training/GPU): preprocess frame → count white pixels per slot → free vs occupied.
Slot coordinates stored in a pickle file CarParkPos created by a simple picker tool.
FastAPI backend publishes counts; ultra-simple frontend prints one line (e.g., Lot 22(A): 13/150).

Tech Stack

Backend: Python, FastAPI, OpenCV, NumPy
Frontend: Plain HTML/JS (optional Leaflet badge UI later)
Runtime: WSL2 Ubuntu (works on native Windows/macOS/Linux too)

Repo Layout

ParkAssist/
├─ backend/
│  └─ main.py               # FastAPI server, processes video → counts → JSON + /summary
├─ frontend/
│  ├─ simple.html           # Shows "Lot 22(A): X/Y"
│  └─ (index.html)          # (optional) Leaflet map with badges
├─ media/
│  └─ carPark.mp4           # Sample video feed (replace with campus feed)
├─ CarParkPos               # Pickled list of (x,y) slot tops (made by picker)
├─ main.py                  # Original demo pipeline from the OpenCV repo
└─ ParkingSpacePicker.py    # Click to create/update CarParkPos

Quick Start

0) System deps (video + GUI safety on WSL2)

sudo apt update
sudo apt install -y libgl1 libglib2.0-0 libsm6 libxext6 libxrender1 ffmpeg libgtk-3-0

1) Python env + deps

python3 -m venv .venv
source .venv/bin/activate
python -m pip install --upgrade pip
pip install fastapi "uvicorn[standard]" opencv-python numpy

2) Generate parking slots (if `CarParkPos` missing/outdated)

python ParkingSpacePicker.py
# Left-click to add slot top-left; Right-click to remove; auto-saves to CarParkPos

3) Run the backend (from repo root)

python -m uvicorn backend.main:app --reload --port 8000
# Endpoints:
#   GET /api/lots       -> [{"id","name","free","total","center","updatedAt"}]
#   GET /api/lots/22A   -> {"id","name","free","total","center","updatedAt"}
#   GET /summary        -> "Lot 22(A): X/Y"

4) Run the simple frontend

cd frontend
python -m http.server 5173
# open http://localhost:5173/simple.html

Key Constants (tune as needed)

WIDTH=107, HEIGHT=48 — slot rectangle size (match your picker)
THRESHOLD=900 — pixel threshold separating free vs occupied
Preprocess in backend mirrors original:
- Gray → GaussianBlur(3×3, σ=1) → AdaptiveThreshold(25,16, INV) → MedianBlur(5) → Dilate(3×3, iter=1)

Troubleshooting

cv2.imshow fails on WSL: run backend only (no GUI needed) or install GTK and use opencv-python (not headless). The frontend shows the counts.
Video won’t open: verify media/carPark.mp4 exists and path resolution (use pathlib if running from backend/).
Wrong counts: re-click slots with ParkingSpacePicker.py and/or adjust THRESHOLD.

Roadmap

Multiple lots (LOTS dict in backend) • polygon masks per slot • Leaflet map badges • health check (/healthz) • better error handling for video feeds • export historical CSV.

9/27/2025