Inside and outside driving situation monitoring using cameras
Cygnify's AI technology monitors and predicts both the situation outside the vehicle, in terms of current driving context and possible hazards; and the situation inside the vehicle, including driver fatigue, distraction, and other driver states.
The most important sensor inputs come from small, relatively inexpensive and unobtrusive outward-facing and inward-facing cameras. These can be mounted inside any vehicle cabin. These sensor inputs are processed by a combination of state-of-the-art deep learning-powered computer vision algorithms, which extract many relevant pieces of information for the next levels of processing.
Integration and forecasting for proactive ADAS information
In the next levels of processing, information is aggregated, and proprietary (again deep learning-powered) forecasting is run in order to not only interpret the situation, but also to predict the likely situations ahead. Those situations can be potentially hazardous "external" situations encountered by the vehicle due to complex traffic situations ahead or other road users' behavior; but also "internal", driver state-related situations, e.g. increasing driver fatigue, gradually reaching the level of serious degraded driving performance.
By integrating these multiple sources of information and predictions, both from the outside and the inside of the vehicle, Cygnify's technology is uniquely positioned to give valuable proactive warning information, countermeasure information, and auxiliary data to the vehicle's driving assistances systems, (partially) automated driving systems, and their corresponding human-machine interfaces.
Onboard computing, combined with offboard training
The monitoring and prediction algorithms run, in real time, on an efficient, powerful, and rugged off-the-shelf onboard computer, which can be powered using the vehicle's battery power. The AI models encapsulated in that real-time software are trained offboard on Cygnify's high-performance private cloud servers. Through automated wifi or sim-card enabled data connections, models can be automatically synchronized and relevant vehicle and driver data uploaded.
Large amounts of driving and driver data, much of it proprietary and representing many kinds of relevant driving and driver situations, are used in the training process. For both algorithm development and datasets, we partner with some of the world's most respected research institutes and industrial partners, such that we have access to large and unique datasets on driver fatigue, driver distraction, naturalistic driving data containing many hazardous driving situations, etcetera. These are key ingredients to Cygnify's model accuracy.
Cygnify is currently actively working towards deploying the first commercial products, in collaboration with several world-leading automotive and other industry partners active in transportation and safety systems.