Distracted driving is a HUGE problem all over the world, accounting for a staggeringly high number of deaths every year. This, combined with personal experiences some of us had, inspired us to develop a platform that tries to deincentivize this kind of behavior. Endrava not only alerts drivers when they are engaging in distracted behavior, but also provides an economic incentive for them, and their insurance providers to stop distracted driving.
What it does
Endrava uses machine learning and computer vision algorithms to track your gaze using a webcam, alerting you when you get drowsy, or start looking down at your phone. It then aggregates your gaze data into the cloud and uses DocuSign to create an initial agreement between you, the driver, your insurance provider, and your employer. Then your insurance provider and employer (a truck driving company, for example), can reward you for driving distraction-free.
How we built it
We used a graze-tracking machine learning algorithm to identify times when a driver was getting drowsy or distracted by looking at their phone. We worked to make the algorithm fit our specific use case, which combined drowsiness and mobile device distraction.
We then created a continuous data ingestion mechanism in Python, which handled the uploading of distraction events into a MongoDB database located on a Google Cloud Compute engine.
Following this, we created a Node.JS REST API, which we integrated with several DocuSign APIs, to generate dynamic contracts based on a driver's details. We built a frontend using React in order to display a list of _ distraction reports _ to the driver, who could then download these documents and view the number of distractions they had, and the resulting changes to their insurance rates.
Challenges we ran into
We first started with barebones OpenCV to track eye movement and gaze, however, we soon realized it was an unnecessarily difficult and slow approach. We then pivoted to use a well-tested machine learning gaze-tracking algorithm, and found that this proved effective in detecting drowsiness and distracted driving.
We also had trouble initially authenticating with the DocuSign API, however, we were able to fix this eventually by working with the DocuSign mentors.
Accomplishments that we're proud of
We're proud of making a hack that combines hardware, computer vision, and machine learning, areas that we all find interesting, but haven't had much experience in. We're also proud of making a system that has so many moving parts: computer vision and ML, cloud technology, the DocuSign API (which is extremely complex), and a REST API deployed on cloud architecture.
What we learned
What's next for Endrava
We would like to continue improving Endrava's distraction detection algorithm, and make a more comprehensive API that leverages more of the DocuSign to deliver a complete experience to all parties (the driver, insurance company, and the employer).