LUX magazine | posted: 12June 2019

SCIENTISTS have developed an artificial intelligence algorithm which can trick office workers into believing that all the building’s lights are illuminated – when they’re not.
 

The University of Verona brought in expert knowledge and design skills for estimating the gaze of people and their future motion with deep neural network model.

The European Union-funded project has successfully combined advanced research in computer vision and modern technology to develop a lighting control paradigm in which each person in the office perceives the entire office as ‘all lit’, while lights, which are not visible, are switched off.

The software estimates the light propagation in offices in real-time and computes how much of it is perceived by the people. Then smart lighting control adjusts the lighting autonomously based on the presence of people and on their position within the office. The concept works particularly well in large open-plan offices, as farther-away luminaires may be dimmed without altering the comfort of employees and their sense of security.

The four-year project – dubbed SceneUnderLight – is a collaboration between Osram, the University of Verona and the Italian Institute of Technology (IIT) in Genova. Its remit was to come up with a method of delivering ‘maximum comfort and sense of security while granting large savings in energy consumption’.

The participants estimate that technology using this AI lighting control could expect to see savings in energy consumptions of up to 65 per cent.

Since 2015, the project members have targeted two main goals: the understanding of the light and scene structure in an office as well as the understanding of the human factor in an illuminated scene.

IIT provided unique expertise in the 3D light estimation in large and complex office scenes, from colour and depth images provided by a modern RGB-D camera while the University of Verona brought in expert knowledge and design skills for estimating the gaze of people and their future motion with deep neural network model.

Fabio Galasso, head of the computer vision R&D activities at Osram, brought together the team consisting of Marco Cristani, heading the department of Vision, Processing and Sound of University of Verona, and Alessio Del Bue, head of the Visual Geometry and Modelling Lab of IIT.

SceneUnderLight was funded by the EU within the Horizon 2020 framework as a Marie SkÅ‚odowska- Curie Actions Project. It’s not clear at this stage if Osram plans to commercialise the technology.