Track: Clean up your city
Try it out link: located in the slide deck below
About the project:
Github:
Inspiration
Urban environments face ongoing issues with pollution and neglected infrastructure, which can negatively impact public health, the environment, and city aesthetics. Inspired by the need for more efficient and proactive environmental monitoring, we envisioned a solution that integrates artificial intelligence (AI) to help detect and classify pollution in real-time. By leveraging computer vision and YOLO-based deep learning models, we aimed to create a tool that would empower city workers, volunteers, and community members to make quicker, more informed decisions regarding urban cleanup efforts.
What it does
GeoClean SJ is an AI-powered web application designed to detect and classify six types of urban pollution: birds, roadkill, trash, graffiti, potholes, and weeds. The system allows users to upload images, which are automatically analyzed using the YOLO deep learning model. Through image processing, the app delivers real-time pollution predictions. GeoClean SJ also integrates with existing city surveillance infrastructure, enabling automatic detection and reporting of environmental issues, making city maintenance efforts faster, smarter, and more efficient.
How we built it
We built GeoClean SJ using a combination of web development and deep learning technologies. The core of the system is a YOLO-based model trained to detect urban pollution categories in real-time. For the frontend, we used Streamlit to create an easy-to-use interface for image uploads and instant results. The backend processes the images with the YOLO model. We also integrated with city surveillance cameras and patrol vehicles, allowing data to flow seamlessly into the system for automated environmental monitoring and reporting.
Challenges we ran into One of the biggest challenges was ensuring the YOLO model was accurately detecting and classifying the six pollution categories, especially with varying lighting conditions and image quality. Training the model with a diverse dataset took time and required significant fine-tuning. Additionally, integrating with city infrastructure presented logistical challenges, such as establishing secure data streams and ensuring the system could handle a large influx of image data from surveillance cameras. Balancing real-time performance with accuracy was also a key hurdle.
Accomplishments that we’re proud of
We’re proud of building an AI-powered system that is not only functional but also scalable and user-friendly. The real-time image classification works seamlessly, making it possible for cities to detect and address urban pollution more proactively. We also integrated GeoClean SJ with surveillance infrastructure, making it easier to track and report environmental issues automatically. Despite the challenges, we’ve created a tool that can make a real difference in maintaining cleaner, healthier cities.
What we learned
Throughout this project, we learned a lot about the complexities of AI in real-world applications. Training the YOLO model for such a specific use case required a great deal of data curation and fine-tuning. We also learned about the challenges of working with city infrastructure and the importance of seamless integration between new technology and existing systems. On the software side, we gained valuable experience in web development, machine learning, and creating user-friendly interfaces for complex AI systems.
What’s next for GeoClean SJ
Looking ahead, we plan to expand GeoClean SJ to support additional pollution categories and improve its detection accuracy with more diverse training data. We aim to integrate with more surveillance infrastructure across the city to broaden the system's coverage. Additionally, we plan to introduce real-time alerts and dashboards for city planners and environmental agencies to make quicker decisions. As the system scales, we hope to partner with other cities to help improve urban environments worldwide.
Log in or sign up for Devpost to join the conversation.