-
Screenshot of the Dashboard
Problem Definition
Every year, thousands of Canadians succumb to fatal and personal injury vehicular collisions on public roads. In 2018, the number of fatalities due to road collisions reached 1922, a 3.6% increase from 2017’s number of fatalities. This further signifies a growing need for a practical solution that aims to prevent the loss of life in a serious vehicle. More specifically, ensuring instant live-reporting of road accidents could potentially diminish the number of annual fatalities.
Inspiration
The inspiration of RoadSense derived upon discovering emerging technologies that interconnect city-traffic infrastructures and using that to fight against the rising number of fatalities due to vehicular collisions.
What it does
RoadSense, an extension of GE Current’s CityIQ, utilizes machine learning and real-time audio feedback to recognize the occurrence of vehicular collisions near the CityIQ device. It would then pinpoint the location for an immediate emergency response without the need for an individual to dial 911. Afterwards, the data is relayed to the RoadSense database to be displayed on the RoadSense dashboard and finally, a call is placed to 911 with information regarding the accident. Functionality is also added allowing us to pinpoint other crimes, such as gunshots or glass breaking.
How we built it
First, we used Current's CityIQ API and skicit-learn to compare and analyze the data that is being received. We stored data such as; location (latitude, longitude), the time elapsed since the incident, and other components in the Firebase database. We then used Flask and set up Twilio (a cloud communications platform) so that we could programmatically receive phone calls when another accident occurs. The accidents would then appear as an icon on our Google Maps that is in the frontend where we utilized React.js for the overall layout and functionality.
Challenges we ran into
The primary challenge that took a long process to overcome was the lack of data to complement the machine-learning component of RoadSense. Most of this data either had a high cost or required approval by an individual at an organization (eg. Mivia). This was overcome by simply extracting audio clips from popular websites that contain crashes and gunshots. With that, this process required a fair bit of work and only yielded a small dataset (approx 200 points).
Likewise, the API linking us to the CityIQ wasn't perfect and at times wouldn't yield any results when doing a basic API call. This can be easily overcome overtime by having CityIQ make the API more reliable.
Accomplishments that we're proud of
We are proud of our ability to combine IoT and machine-learning technologies to provide a practical solution to enhancing the interconnectedness of cities.
What we learned
It was our first time using scikit-learn for machine-learning models to merge it with CityIQ's IoT platform.
What's next for RoadSense
We plan to take on several steps to ultimately improve the functionality and practicality of RoadSense. More specifically, we intend for RoadSense to detect not only vehicular collisions but also criminal-related noises such as gun-firing and robberies involved with some forms of sound transmission (ex. breaking of windows). In addition to audio processing, we plan to integrate video capture by RoadSense at the time of the incident to provide a greater depth of data related to the incident.
Log in or sign up for Devpost to join the conversation.