Inspiration
In an effort to reduce air pollution in Ho Chi Minh City, authorities have proposed the Vehicle Emission Control Project. One of the proposed methods is to declare Low Emission Zones (LEZ), where only vehicles meeting proper emission standards may pass. Our project starts from the idea that there must be an efficient, cost-effective way to enforce this control method. As the city already possesses a widespread Electronic Toll Collection (VETC) infrastructure, we realized the answer wasn't building new hardware, but connecting existing data silos.
What it does
Our project is a "Virtual LEZ" management system. It acts as a real-time data bridge between streaming VETC RFID pings and the static National Vehicle Registry database. When a vehicle passes a checkpoint, the system instantly cross-references its license plate to check its emission standard (e.g., Euro 4 or below) and engine type. If it violates the zone's rules, the system automatically flags the vehicle and generates a secure violation report, transmitting the data directly to the Traffic Police (CSGT) for official enforcement. To ensure total compliance and prevent evasion, we also integrated an AI-powered Automatic License Plate Recognition (ALPR) module that reads license plates directly from traffic camera feeds.
How we built it
We focused heavily on building a robust data pipeline and integrating machine learning. We created mock datasets to simulate the massive volume of the Vehicle Registry and the real-time VETC streaming data. Using Python, we built the core rule engine to handle data matching and automated violation flagging. For the frontend, we developed an interactive dashboard to visualize traffic flows and emission compliance rates. Finally, we deployed an optimized Automatic License Plate Recognition (ALPR) pipeline to process video frames from traffic cameras, treating the visual data as a secondary input stream to identify un-tagged vehicles.
Challenges we ran into
One of our biggest hurdles was designing a solution that addresses a tangible business and policy problem, rather than just building a technical feature. Simulating a massive, city-wide infrastructure and designing a data architecture that efficiently joins high-velocity streaming pings with a static database in real-time required significant optimization. Additionally, fine-tuning the ALPR model to accurately read license plates from fast-moving traffic frames under time pressure was a tough technical challenge.
Accomplishments that we're proud of
We are incredibly proud of delivering a highly practical, high-impact solution within a tight timeframe. We built a working Proof of Concept that seamlessly combines a robust data pipeline, an AI-driven license plate recognition fallback, and an intuitive data visualization dashboard. We proved that complex smart city initiatives can be achieved through clever software integration and data mining, rather than expensive physical deployments.
What we learned
We learned that the most powerful technical solutions often lie in leveraging existing infrastructure rather than reinventing the wheel. We gained intense, hands-on experience in rapid prototyping, handling real-time data workflows, and translating complex technical architectures into clear business and environmental value.
What's next for Virtual-LEZ
We plan to refine our License Plate Recognition model to handle more complex scenarios, such as nighttime tracking and severe weather conditions. We also aim to implement predictive analytics to forecast pollution hotspots based on historical traffic data, allowing city authorities and the Traffic Police to dynamically adjust LEZ boundaries and optimize enforcement deployment during peak congestion hours.
Log in or sign up for Devpost to join the conversation.