HemenPark combines AI-powered cameras, crowd-sourced reports, and public APIs to show real-time parking availability.
Drivers spend an average of 10–15 minutes each day searching for a parking spot. Wasted time, unnecessary fuel burn, and added stress. Existing solutions either rely on outdated static data or cover only a fraction of real-world spots.
HemenPark tackles this with three complementary data sources: AI-powered cameras that monitor entrances in real time, crowd-sourced spot reports verified by on-device ML and Gemini AI, and edge sensor devices deployed at fixed lots. All three feed into a single live map delivered to the mobile app.
Drive around, guess, waste time and fuel.
Open app, see spots, go straight there.
From live maps to a token economy, HemenPark is a full parking ecosystem.
Parking spots update in real time via GraphQL queries and push-driven refreshes. Color-coded markers reflect occupied/free status within seconds. High-performance map experience built on Google Maps SDK.
Camera frames are sampled and processed in two stages. First, a vehicle detector extracts a bounding box around every car in the frame. Then EfficientNet-B0 classifies each predefined parking region as empty or occupied. Results are pushed to the backend over HTTP REST. No raw video is stored or transmitted.
Users earn HemenPark Tokens (PT) for every validated parking report. Tokens unlock detailed spot info or short-TTL reservations. Dynamic pricing adjusts token cost based on report freshness and demand.
Users report empty parking spots using their phone camera and GPS. Before any upload, a lightweight MobileNet classifier running fully on-device verifies the photo as valid or invalid, with no server round-trip needed. The same MobileNet runs natively on both Android and iOS. Photos are stored in MinIO object storage and reviewed by Gemini 2.5 Pro for verification. Only valid reports influence the map.
Cities and private parking operators can integrate via REST API. The Public API Connector app normalizes data from different sources and merges them into a unified stream.
HemenPark
Shared business logic via Kotlin Multiplatform with Compose Multiplatform UI running natively on both Android and iOS. Swift is used only for platform bridges (TFLite, Maps, notifications). One codebase, two fully native experiences.
Concerts, matches, road closures, and construction all change parking conditions before they hit traditional data sources. Gemini 2.5 Pro scans news and city feeds daily to discover parking-impacting events, and FCM/APNs deliver geofenced push notifications only to users near the affected area. Affected zones are highlighted live on the map so drivers can route around them.
HemenPark aggregates from three distinct sources simultaneously: (1) AI cameras whose frames flow through a two-stage pipeline: a vehicle detector extracts bounding boxes, then EfficientNet-B0 classifies each parking region as empty or occupied. The pipeline runs on backend Celery workers by default; optional Raspberry Pi edge agents can run the same models locally and push results via HTTP REST. (2) Public parking APIs for city and private lot provider feeds. (3) Crowd-sourced reports with user-submitted photos pre-classified on-device by a MobileNet classifier (Android + iOS) before upload.
The Django + Strawberry GraphQL backend applies PostGIS geospatial deduplication, confidence scoring, and stale-data cleanup across all sources to produce a unified, ranked list of "spot candidates."
Dynamic Pricing Engine sets token costs based on report age and demand. OSRM calculates server-side walking ETAs. Gemini 2.5 Pro discovers parking-impacting events daily and verifies crowd-sourced reports. FCM/APNs push notifications keep users informed in real time.
The Kotlin Multiplatform app (Android + iOS) with Compose Multiplatform UI polls spot data via Apollo GraphQL and shows live updates on an interactive Google Maps view. Users earn tokens for validated reports and can reserve spots with short-TTL holds.
Six screens that show how drivers find, reserve, report and earn, all from one pocket-sized experience.
Browse the otoparks closest to you with up-to-the-minute availability, walking distance and live fill bars. Filter by camera coverage, paid/free, or only-available with a tap.
Green, red and amber pins update as cameras, edge sensors and crowd reports flow in. Pan, zoom and tap any pin to see live status, source confidence and reserve in one motion.
Pick a day, dial in your arrival time, set duration. Short-TTL holds keep the spot warm while you drive, with payment handled at the lot for partnered providers.
Spotted an empty curb? Tap, shoot, send. An on-device MobileNet classifier pre-validates the frame so only useful reports leave your phone, and tokens land in your wallet when it is verified.
Top up tokens, unlock packs or jump to a monthly plan for unlimited reservations and premium features. Every spend and reward is tracked transparently in your wallet.
Gemini 2.5 Pro scans city calendars daily and pushes alerts when a nearby event is likely to fill every spot around it. You see the risk before you leave, not after you arrive.
Microservice-first platform with end-to-end security, geospatial intelligence, and edge AI.
Every layer is built with open-source, battle-tested technologies.
A senior capstone team from Middle East Technical University, Department of Computer Engineering.
HemenPark is a scalable smart parking platform built for cities and campuses. Join the future of parking.
Discover HemenPark ↗