This project is divided in two parts:
1- A Convolutional Neural Network (CNN) trained on a semi-supervised fashion to automatically detect outliers on the trajections of the city of Rio de Janeiro. The models were trained using Caffe framework.
2- An online vizualization tool that help users identify and understand outlier buses (outliers) in the bus routes of Rio de Janeiro. These outliers can be spatial -- e.g. buses running outside their average route (far from where they are likely to be) -- and temporal -- e.g. delayed buses.
In order to run the visualization tool you only need to run a local python server (python -m SimpleHTTPServer). This repository already contain some sample data. However, the original data can be obtained in a stream (daily) fashion here or here.
A more detailed explanation please refer to the project description. You can try out the live demo and
see in the video to how interact with the tool.