safe driving
DEEP LEARNING FOR SAFER ROUTES ON THE ROAD
“HOW MIGHT WE USE TECHNOLOGY TO TAKE CARE OF DRIVERS WHEN THEIR INSURANCE COMPANY CANNOT BE PHYSICALLY PRESENT? ”
CASE
An insurance company asked us to use technology advances to increase safety in driving. We had to conceptualize and design a revolutionary concept for safe driving based on Deep Learning technology.
The goal was to reduce the number of accidents and incidents on the road by accompanying and guiding drivers.
SOLUTION
Safe Driving assistant is a navigation extension. It detects and estimates dangerous situations on the road based on accident data, traffic condition, driver profile, status of the vehicle and other circumstances when driving. The alerts and advices keep the driver safe.
CONCEPT
Safe Driving combines data from both internal and external sources. It feeds on data from sources concerning the driver, the vehicle and the traffic.
Road hazards are estimated based on the analysis of road accidents data. Their conditions are nurture the deep learning algorithm, that determines to who, when and where they are most likely to occur. This analysis is combined with real-time data on traffic and road (external), vehicle status (internal) and driver (personal) status to detect on-route risks. The assistant warns the driver of the danger while driving when the risk levels are overpassed. Real-time alerts based on location work along with voice recommendations and/or autonomous solutions to avoid or mitigate possible effects of such hazards.
PROCESS
Data scouting
Due to the large amount of available databases, an initial research was essential to identify how these data work and could be transformed into useful information for deep learning. Our study focused on understanding how different data sources could be interlinked to contribute to our future solution.
Experience design
Then, we designed the drivers’ experience. We considered the optimal way to use the algorithm information to not only ensure road safety, but also improve users' driving habits.
For this reason, we defined nine possible cases of use of the solution on the road, describing diverse situations in which the solution acts facing meteorological conditions, retentions, emotional changes and states of the vehicle that constitute a danger for the insured. For each of them, we designed a reaction of how the solution alerts and interacts with the driver.
Solution sketch
One of the main challenges of this project was to transmit the mechanics of the solution to the software team to develop it. Several trials, changes and iterations of the solution sketch and report were needed to make sure the algorithm building team would consistently follow the proposed design.