Inspiration
The goal was to create a tool to prevent unnecessary life loss on the national railway network.
What it does
Using machine learning and image classification we identify whether there is a person on the train tracks at a station and notify authorities with the aim of reducing the risk of an accident. using a database of interconnected stations we can keep track of stations that are safe for trains to pass through.
How we built it
AI Model built with Tensorflow using a Convolutional Neural Network. The data was cleaned and prepared in OpenCV. Flask backend loaded the railway network from a SQL database.
Challenges we ran into
Collecting training data for the classification model. We overcame this by building our own 3d model of a train station. We screen-recorded the 3d-model from multiple angles and lightings with different objects in different locations.
Accomplishments that we're proud of
The prediction model thrived even outside the 3d-model. When applied to other images found online it was reliable and showed great promise in real-world data.
Data structures and databases to model multiple train lines allowed for a clean realistic simulation of a train station and this showed in the UI.
What we learned
We as a group have acquired various knowledge through this hackathon. For instance, one of the members learned about HTML and CSS to create the UI of the application from scratch and how it interacts with a database like SQL. Moreover, we also learned about how an AI can process various images and get a result based on image classifications and pattern recognition.
What's next for Project Halo
Using Actual CCTV footage and having access to a bigger more diverse dataset.
Log in or sign up for Devpost to join the conversation.