The most powerful accident prevention solution in the world
We’ve created
an accident prediction model
- Using a new application of the Transformer model that Chat-GPT has made famous, we have created a FULLY LEARNED accident prediction model.
- We can predict seconds in advance when an accident is likely to happen, and where and what kind of incident will occur.
Key Features
Fully Learned Model
Our fully learned model learns to look at subtle combinations of factors rather than a simplistic single-factor view. For example, a driver’s failure to look in a particular direction combined with a pedestrian entering unexpectedly from that side of the scene.
Vision Transformers
Vision transformers allow our model to look at changes over time. Unlike existing Al models in the auto industry, it can infer directions and speeds of objects and use this to predict a future accident.
Driver Embeddings
Driving style is a major contributed to the rate and type of accident. One situation may be totally safe for one driver but dangerous for another. InsureVision’s model continuously learns ‘driver embeddings’ that encode the driver’s style and habits and uses this to inform its predictions.
720 Video
InsureVision combines
interior and exterior
video streams to create a
‘720 degree’ view of the
situation. This is vital for
understanding where
driver inattention
intersects with dangers in
the environment to
accurately predict
accidents.
Next-Generation accident prevention
Its fully learned video transformer model uses modern transformers and vision networks together to learn to predict accidents from video in a radically new way.
As a fully learned model, InsureVision offers far more predictive power than heuristic models that are trained on a few categories of incident (inattention, pedestrian detection, etc.).
Al vision models in the automotive arena are almost exclusively atemporal: they look at one image at a time.