Inspiration
Due to the large amounts of pollution and increasing rate of global warming, we were inspired to create a model to help urban planners reduce congestion time in a city. Furthermore, we all spend a lot of time commuting or being stuck in traffic during rush hours, so being able to predict this congestion time before an area is built can help urban planners improve people's travelling times.
What it does
This project takes in satellite imagery of a city and predicts congestion based on the urban landscaping.
How we built it
We built this project using a convolutional neural network. We obtained satellite imagery as well as traffic data from Google images, and we used this data to train the neural network to recognize patterns in the urban landscape that may increase or decrease congestion in the area.
Challenges we ran into
Challenges we had is how to take millions of screenshot from Google Maps and find a best fit model to train our AI.
Accomplishments that we're proud of
We were able to create a CNN which was able to create correlations between certain urban landscape features and congestion time, with only a limited amount of time and knowledge of ML prior to the start of this hackathon.
What we learned
We learned how to used Tensorflow to create a model and train an AI using a convolutional neural network. As well as how to use Google Maps data to help create our model.
What's next for Using Context-Aware CNN to Model Congestion in Urban Areas
Next steps would be to create a user interface for our model and build relationships between areas of varying size such: - Regression layer to return "intensity" rather than category - 3 photos of differing zoom levels, sharing the same center - Input shape now 9 instead of 3 Generalize feature detection over different cities and landscapes. Add in more channels of satellite imagery such as NIR to aid in landscape detection.
Log in or sign up for Devpost to join the conversation.