Our state-of-the-art AI-enabled infrastructure is based on over 25 years of proven research in the field of driving behavior analysis, traffic accident analysis, distracted driving, road safety, transportation engineering, modelling / big data analytics and machine learning and is backed by a top tier award-winning engineering team.
We implement the most advanced ML techniques to exploit the recorded data. A wide toolkit of ML algorithms has been developed to:
- Detect harsh events and mobile use
- Identify the trip transport mode (car, motorcycle, mass transit)
- Recognize whether the user is a driver or a passenger
- Provide spatiotemporal analysis of the driving data
- Analyse whether user drives in an eco friendly manner
- Detect severe crashes
We use advanced signal processing approaches to remove the noise from the recorded raw data and keep only the components that are actually related with driving behaviour. By doing so, we ensure that all possible data sources such as OBDs, smartphones and in-car telematics, produce equivalent results which can be easily integrated in the OSeven device agnostic platform.
The OSeven event detection algorithms have been calibrated in order to produce equivalent results from any recording device (OBDs, smartphones, connected vehicles) using naturalistic & controlled on road experiments, driving simulator experiments and advanced signal processing methods. Our results have been verified by Road Safety and Driving Behaviour experts from the National Technical University of Athens, who are also certified by the Belgian Road Safety Institute.
MOBILE USE DETECTION (WITH OVER 98% ACCURACY)
Mobile use while driving has become one of the most significant risk factors. We have managed to develop an accurate detection algorithm which measures mobile use, based on data fusion and ML approaches. Since privacy is one of our greatest principles, our algorithm requires no additional permissions to other smartphone functionalities and no access to personal data, but it relies solely on data from the smartphone sensors.
TRANSPORT MODE DETECTION (WITH OVER 85% ACCURACY)
We use several means of transportation (car, motorcycle, bicycle, mass transit) every single day. However, every mean of transportation has its very own pattern. We can accurately detect the user’s mode of transport by utilizing smartphone sensors and our AI-enabled infrastructure.
DRIVER-PASSENGER RECOGNITION (WITH OVER 92% ACCURACY)
One billion insured vehicles, one billion different driving patterns. Everyone has its own driving pattern strongly affected by its personality and daily routine. Additionally, our habits change while we are passengers rather than drivers. Think of yourself while being a passenger. Do you position or use your device the same way as when driving your own car? We have developed a set of ML algorithms that can reliably determine if the user is the driver or a passenger taking into consideration all the above parameters.
The driving score for OSeven is not just a number or a marketing tool, but it reliably quantifies the risk associated with a specific driving behaviour.The main scoring categories are:
Speeding: Measures the distance and time of driving over the speed limit and the exceedance of the speed limit
Distraction: Measures distraction caused by mobile use during driving
Braking: Measures the frequency and intensity of harsh brakes and braking aggressiveness
Acceleration: Measures the frequency and intensity of harsh accelerations and accelerating aggressiveness.
The driving score is calculated using a large number of variables, parameters and a sophisticated algorithm developed by our research and engineering team based on 25 years of research in the fields of driving behaviour analysis, traffic accident analysis, distracted driving, road safety, transportation engineering, modelling / big data statistics and ML.
For the development of our algorithm we are also working closely with the Department of Transportation Planning and Engineering of the School of Civil Engineering of the National Technical University of Athens and the respective NTUA Road Safety Observatory (NRSO).
For every trip a user completes, we record a large amount of data and we produce a large number of metrics and features in order to evaluate driving behaviour. However, we focus on providing users only the valuable information which will help them improve. Therefore, the OSeven Data Science Team has developed algorithms that filter the most critical information, the Highlights of each trip, that are communicated to the users together with a customized message based on their driving behaviour in the specific trip and their overall driving profile and performance.
SEVERE CRASH DETECTION (WITH OVER 99% ACCURACY)
When it comes to road crashes even seconds can make a difference.
2.4% of crash related fatalities could have been prevented if the crash location was determined on time
21.1% crash victims could have survived if they were transported sooner to a hospital or trauma center
Source:European Road Safety Observatory
OSeven has developed a Severe Crash Detection (SCD) algorithm utilizing data from the smartphone sensors and advanced data fusion algorithms. The algorithm has been calibrated and is constantly being optimized based on a large database of crash tests (frontal and side car to car crashes and car to barrier crashes) and the recorded trips in OSeven database. Our SCD algorithm is embedded in the smartphone application and it can distinguish the actual severe crashes from other events that could produce similar signals, (e.g. harsh breaks with high intensity and mobile phone drops), achieving an astounding accuracy of over 99% and zero false positives, in addition to minimum battery consumption.