SafeRoute AI | Devpost

Inspiration

Navigation applications in cities are not about security, but speed. As more people start to be worried about crime, night travel, and collision hot spots in Toronto, we felt the desire to develop a system that would provide people with safe routes. The concept behind the SafeRoute AI was founded upon the integration of real cyclist, traffic, and crime data to develop a tool that will allow pedestrians and students to travel with confidence.

What it does

SafeRoute AI processes three real datasets of Toronto and transforms them into an overall city-wide safety heatmap. It pays off risk scores based on cyclist collisions, traffic accidents, and major crimes and solves a risk-weighted pathfinding problem to suggest the safest walking path between two points not necessarily the shortest.

How we built it

We processed and standardized three datasets and identified important features (injuries, type of crime, timestamps, location) and mapped all incidents onto a 10x10 geospatial grid. To determine the safest path, we provided risk scores to the individual cells with weighted formulas, constructed a grid graph on the risk scores, and applied the Dijkstra algorithm. Lastly, we plotted data in terms of heatmaps and path overlays.

Challenges we ran into

Dealing with irregular column formats and blank data. Accurate conversion of lat/long values to grid cells. The trade off between crime, traffic, and risk to cyclists. Our safe-path algorithm made realistic routes. Integration of three datasets of various structure.

Accomplishments that we're proud of

Effective use of three significant datasets into a single safety model. Creating a complete risk hazard map and sturdy routing engine. Constructing a clear, decipherable system of risk-scoring. Creating fine urban insights out of actual Toronto data. The development of a solution that can actually enhance the safety of people.

What we learned

We also acquired knowledge of how to handle multi-source urban data, use geospatial binning, create graph-based pathfinding algorithms and visualization of complex risk patterns. It covered collaboration, data cleaning, ML feature engineering, and converting raw data into an effective safety tool, as well.

What's next for SafeRoute AI

Add live information (weather, events, live traffic). Create a mobile interface that can be used by normal users. Predict risk in the future by using advanced machine learning. Add personalization (e.g., do not go to unlit places, do not go to late night hotspots). Extend the model to other cities other than Toronto.