-
-
Logo
Inspiration
Due to the rise of violent crimes all across the nation, we wanted to develop an app that would allow us to combat this violence. By leveraging the cumulative eye-witness testimony of many people, this app ensures that all individuals can be sure to live in a safer tomorrow!
What it does
SafeStreets features a dynamic map displaying a heatmap of historical crime data and real-time markers for user-reported incidents. Users can report violent acts, and the app notifies them when they are near a reported incident, allowing them to take immediate precautions. Other features include a relevant news section and automatic image caption generation for faster reports, which is done by leveraging the combination of two different machine learning models.
How we built it
We used React Native for the front end, Python and Flask for the backend, Firestore and Firebase Storage for data storage, and Python with Selenium for web scraping. We've fine-tuned the CLIPxGPT Captioner model to provide accurate image descriptions over 120 epochs on the Flickr30k dataset, consisting of 30,000 images.
Challenges we ran into
Integrating the backend and training model with the app's front end was very difficult. Additionally, web scraping relevant information and finding trusted news sources for the Heat Map's display of information was another challenge.
Accomplishments that we're proud of
We are proud that we used Artificial Intelligence with a mobile application, ensuring accessibility for all individuals. We are also proud of our efficient data transfer techniques; specifically, we are proud that all users can quickly receive live updates from other users. Finally, we are proud of our efficient data scraping techniques from governmental sites, which we know have trusted information.
What we learned
We learned how to integrate a Flask server for a backend machine learning model with the frontend React Native components. Additionally, we learned how we can transfer images between all users (no matter the device) by utilizing Firestore Storage. We also fine-tuned our Google Firebase schools, where we keep a clean record of all of the data that we encompass.
What's next for SafeStreets
Outside of improving the accuracy of the image captioning model, we hope to incorporate not only a visual aspect but an auditory form of report as well to make users able to report incidents faster and more efficiently. Additionally, adding more and more trusted news sources is needed to expand the heat map to even more violent areas across the United States!
Log in or sign up for Devpost to join the conversation.