Schneider Electric wanted to integrate intelligence into an infrared sensor that would allow the recognition of presence and human activity in a building. In addition to having to design algorithms to detect human presence, these had to be embedded in a very low-power sensor.
After dimensioning the project and conducting a feasibility study, Neovision then designed computer vision and machine learning algorithms to detect human presence and activity. These algorithms were then embedded on a low-power microcontroller.
Schneider Electric now has a smart sensor to address the smartbuilding market. These sensors are now able to detect a human presence and to classify it thanks to the algorithms they have. Moreover, by taking other parameters into account, these microcontrollers make it possible to manage energy intelligently and automatically.
Schneider Electric is a world leader and specialist in energy management and automation. The group has 144,000 employees and is present in more than 100 countries.
The collaboration between the multinational company and Neovision began in 2015. The objective shared by the two companies was to develop a technology capable of detecting human presence and activity.
This was a pioneering project for Neovision as it entered the AI market. At the time, nobody was talking about AI, but Neovision already had solid and much sought-after skills. Indeed, we could think that a group the size of Schneider Electric could carry out this project internally, but the technical constraints represented a major challenge and only high-level AI experts could solve it.
Knowing Schneider Electric’s market, this project made perfect sense. The final goal was to be able to combine energy and automation. To do this, the Schneider teams had designed a new very low-power product coupling an infrared sensor to a microcontroller, but which now needed to be made smart. By smart, we mean making it capable of detecting and recognizing human activity in a building. It could then feed information back to an energy management system in order to manage energy in the best possible way based on the people present in the building and their activities.
Gilles Chabanis, head of the Smart Sensor department, and Dominique Persegol, head of the MIRTIC collaborative project that gave birth to this sensor, called on Neovision to carry out this R&D project. It must be said that they knew who to turn to: indeed, Lucas Nacsa, Neovision’s CEO, had had the opportunity to work with them during his experience at Inria.
First, Neovision carried out a feasibility study. Of course, it had to take into account the significant technical constraints imposed by the microcontroller, particularly in terms of memory and calculation.
Then, Neovision isolated the solution that seemed to be the most appropriate and efficient. And Neovision began designing the computer vision and machine learning algorithms. They had to integrate and take into account a lot of information, including: occupancy rate, location of people, and measurement of average surface temperatures. This information was necessary to control the building’s air conditioning.
The main challenge lay in optimizing the memory and algorithm calculations. Indeed, the software had to be embedded and run easily on the microcontroller. This we managed to do thanks to our knowledge in applied mathematics and embedded computing.
Finally, we were interested in posture recognition (sitting, standing or lying down). The aim was to take into account the activity and comfort of the people
The project is a success because it is currently in its large-scale commercialization phase at Schneider Electric.
COMPUTER VISION, MACHINE LEARNING, EMBEDDED AI
« Neovision was able to quickly take into account the many technical constraints of this project. We appreciated the pragmatic and reactive aspects of Neovision as well as their varied skills in the field of algorithmics, which allowed us to explore and test various solutions efficiently. »
Dominique Persegol, Project Manager at Schneider Electric
16 October 2020
Computer Vision, IA embarquée, Machine learning