Inspiration
Some cyclists are reckless, others are extremely careful, yet, some road accidents are unavoidable to all. With the number of frequent cyclists in Montreal rising, there is now an incentive for cyclists to seek protection in the form of insurance. Would it be possible to compute the inherent danger a cyclist runs when riding a given route? We from BIXI dangers are keen BIXI users, and have always wanted to analyze data from our BIXI profile. This project gave us an excellent opportunity to combine traffic incidents with BIXI data.
What it does
Our project aims to help cyclists understand the danger they run by biking passed historical high-collision intersections. We combine BIXI data with collision maps to optimize safety and quantify the risk of collision. Our analysis provides insight on how to reduce risk accident for cyclists, bike-share companies, insurance companies and municipalities alike.
How we built it
- We obtain BIXI user data in PDF format. We parsed start and end point of each ride into CSV files.
- We estimate the path taken by computing a database of shortest routes between BIXI stations using the Open Street Map (OSMNX) package and retrieve the path based on start and end BIXI station.
- We then computed a collision map using total yearly road accidents data and used unsupervised learning methods (DBSCAN, Gaussian Mixture Models) to obtain a collision risk map.
- We convolve the traffic map with a collision map to assess the absolute collision risk allowing BIXI to predict the amount of damage they should expect in a year.
- We renormalized the collision map by bike traffic density to assess a danger index for each user, allowing for the design of personalized insurance policies for BIXI users based on their danger profile.
Challenges we ran into
On the processing side, turning PDF files with BIXI data into usable CSV files, creating a graph with OSMNX and calculating the shortest paths between all BIXI stations, despite the high computational effort. On the frontend, we had difficulties transferring our datafiles across platforms.
Accomplishments that we're proud of
We were able to turn an elaborate idea into a close-to-finished product in a short timespan. We managed to compute danger indices and danger profiles that accurately reflect the ground truth. One of our teammates is a known reckless cyclist (for one, he does not change his route based on the presence or absence of bike lanes). Our index reflects his risky riding.
What we learned
St. Denis is a dangerous street! Don’t go to Berri / Ontario! Shortest path taken is computationally painful! Parsing PDFs is hard! Aim big, but start small!
About the last point, we were thinking too much about the demo and too little about the analysis at first. We wanted to display the result through a web application using Google API, which was not feasible. We then shifted our focus from the frontend to be able to finish the analysis.
What's next for BIXI Dangers
We will first finish the processing pipeline, expand our route database to include the full island of Montreal, and finish our interactive website. Soon, a user can upload BIXI and GPS data, and compute his or her danger profile.
After this, we intend to contact BIXI to discuss our approach on computing danger indices from public data, and who knows, we can grab the interest of insurance companies and municipalities with our analysis.
Finally, we can refine our analysis by including the time of the day, month of the year and ride duration, and improve the algorithm for computing the danger index.
Log in or sign up for Devpost to join the conversation.