Understanding how people use indoor spaces is increasingly important for creating buildings that are energy efficient, comfortable, and responsive. Almende and Signify have been working on new ways to detect human presence and activity indoors using wireless signals and AI. Instead of relying on cameras or intrusive sensors that require people to carry tags or wearable devices, our approach uses small changes in radio signals that naturally occur when people move through a space.
The Living Lab
To develop and test this technology, we created a long-term experimental environment in our own office. This setup, which we call the Living Lab, has been operating for several years and allows us to study wireless sensing in a realistic everyday environment. The Living Lab was developed jointly with Signify. In this setup, we combined our Crownstone smart-plug infrastructure with wireless sensing technology installed by Signify.
The result is a dense network of small devices distributed throughout the office that continuously measure signal interactions within the space. By combining these different sensing technologies, we can observe how wireless signals behave when people move, work, or pass through the environment. This real-world setup allows us to collect large amounts of data and test new algorithms under realistic conditions rather than in controlled laboratory environments.
Detecting presence from wireless signals
Wireless signals behave differently when a person is present in a room. Human bodies absorb and reflect radio waves, which causes small fluctuations in signal strength between devices. In our Living Lab, multiple Crownstone smart-plugs exchange wireless signals and measure the Received Signal Strength Indicator (RSSI) between them. These signal measurements fluctuate continuously due to environmental conditions, furniture, and movement in the room. At first glance, these fluctuations look like random noise. However, when large amounts of data are collected and analyzed with machine learning techniques, patterns begin to emerge. Certain signal patterns correlate with human presence or movement in the environment. To improve the reliability of this detection, we also collect additional contextual information from environmental sensors, such as temperature and humidity. By combining these different sources of data, the system becomes better at distinguishing between normal signal variation and changes caused by people moving through the space.
AI at the Edge
A key aspect of this project is that the intelligence is distributed throughout the system. Instead of sending all raw sensor data to a central server, parts of the AI processing run directly on the sensing devices themselves. This approach, known as Edge AI, allows the sensors to process signal data locally and extract useful information before sharing results with a central hub.
Running AI locally has several advantages:
- It reduces network traffic
- Improves response time
- Helps preserve privacy because raw sensor data does not need to be transmitted or stored centrally
The Crownstone smart-plugs run a lightweight AI model that analyze signal fluctuations and detect patterns associated with human presence. More complex processing and sensor fusion are handled by a local hub, which combines information from multiple devices to build a spatial understanding of the environment.
From room-level detection to indoor localization
Current Crownstone-based systems already support room-level detection. This means that the system can determine which room a device is located in by comparing signal measurements with previously recorded signal patterns.
The work in the Living Lab aims to extend this capability further. Instead of only determining whether someone is in a room, we want to determine where within the room activity is occurring. To achieve this, we are exploring a multi-model localization approach. Multiple sensing methods analyze wireless signals from different perspectives and combine their results to build a more detailed spatial model of the environment. This allows the system to estimate positions more precisely within a room.
In the future, this approach may also enable detection of movement direction and activity patterns, providing deeper insights into how spaces are used.
Data collection and analysis
The sensors in the Living Lab continuously collect wireless signal measurements and environmental data. These data streams are stored and analyzed using monitoring tools that allow us to observe signal behavior over time. By studying these datasets, we can train and evaluate new AI models that identify patterns related to presence and movement. The system learns to distinguish between empty-room conditions and situations where people are present, even in environments where wireless signals are naturally noisy. Early results from the experiments show promising differences between these scenarios, indicating that wireless sensing combined with AI can reliably detect human presence.
Building a scalable and privacy-friendly system
An important goal of this research is to develop sensing systems that are practical for real-world edge settings. This means addressing not only technical performance but also broader requirements such as scalability, energy efficiency, resilience, and privacy. The distributed architecture used in the Living Lab supports these goals. Edge AI allows most data processing to happen locally, while federated and distributed learning techniques make it possible to improve models without centralizing sensitive data. By combining multiple sensing technologies, AI models, and distributed processing, we aim to build systems that can scale to larger environments while maintaining reliability and privacy.