Some people think fully self-driving cars are the best thing since sliced bread, while others see them as a disaster waiting to happen. Whatever side of the fence you may be on, fully automated cars are likely to become a reality in the near future, with various levels of automation already on the road. That being the case, technologies to watch over these systems and identify potential problems before disaster strikes are very much needed.
The Fraunhofer Institute for Optronics is actively developing one such technology. hey have designed a system that can detect the activity of the driver. While there are other technologies in existence that can detect driver activity, they tend to be focused on fatigue detection. The Fraunhofer Institute’s device, on the other hand, was designed to detect various activities, and then to draw conclusions from that data. They are able to, for example, understand how long it might take for a driver to take back control of the vehicle, depending on what they are doing.
In addition to identifying facial features, the system is also able to detect the driver’s current pose by computing a digital version of their skeleton. This body position data is fed into artificial intelligence algorithms which can then infer what that person is actively doing. This information can tell the computer whether someone is sleeping, looking at the road, how distracted they may be, and how long it would take them to return their focus to the road.
The system has been designed in such a way that the camera can be placed in any number of locations, giving interior designers a great deal of freedom. Collaborations are already underway with automakers such as Audi and Volkswagen, so perhaps this technology will find a way into a car near you in the future.
If you find having a camera in your car that is always pointed at you distasteful, it may help to know that no video is recorded or transferred over a network. All camera data is analyzed in real time, and not stored in any way. Furthermore, the machine learning models do not need personalization, so no personal data needs to be collected for initial training purposes.