Fujitsu Advances AI for Social Innovation, Aiding Safer Driving with New Deep Learning Solution

Fujitsu Laboratories of Europe has unveiled its next generation Artificial Intelligence (AI) solution, which has been designed to provide real-time insights from a variety of social signals. This new social innovation solution represents a significant breakthrough in terms of applying a machine learning analogue to achieve a natural and visual approach to data classification and analysis. An initial application on show for the first time at the Fujitsu Innovation Gathering in Paris involves aiding safer driving, enabling insurance companies and fleet operators to assess driver behavior in real time.

Fujitsu’s new AI solution considerably extends conventional sensor-based activity recognition technology, which is traditionally geared towards health and personal fitness applications. In the safer driving use case, Fujitsu Laboratories of Europe has developed an analytics solution to analyse driver behaviour-based data, captured from a single wrist-worn accelerometer. Rather than recording activities such as walking, running, and cycling, it focuses on detecting ‘micro-interactions’ such as smoking, drinking, or using a mobile phone. These tools provide a much wider range of capabilities for the real-time analysis of driver awareness and the factors affecting their attention.

Fujitsu is actively conducting R&D that advances the concept of Human Centric Innovation, enabling a hyperconnected world from back-end ICT infrastructure to front-end Internet of Things (IoT) devices. Fujitsu Laboratories of Europe’s Executive Vice President, Dr Adel Rouz, expands : "Our new social innovation solution is an example of how we can harness data from connected devices and apply deep learning to unlock key information, providing real-time analysis and insight for time-sensitive applications, such as the safer driving use case. We are confident that it has value for a wide range of applications, including energy management for example. Using a variety of smart meter sensors, measuring factors such as CO2, humidity and temperature, to infer room occupancy, our solution can identify with high accuracy whether or not a room is occupied by analyzing changes in sensor measurements. Another area is the financial services domain, where the technology can be deployed for searching and analyzing trends in financial data, represented as time series signals and analysed with the same ‘imagification’-based deep learning technology. The objective is to enable analyists instantly to assess the state of the market through real-time analysis of thousands of similar situations in the past."

Fujitsu’s solution overcomes the inefficiencies of existing machine learning approaches by enabling a more natural, human-friendly visual approach to the task of data analysis. Rather than making predictions on raw numerical data, the system makes predictions based on visual representations of the same data, leveraging the high accuracy of state-of-the-art deep learning based recognition systems.