Introduction
Wi-Fi Sensing with Deep Learning is a newly launched and fascinating concept that harnesses the power of Wi-Fi signals to detect and interpret human movements and activities within a given space.
Testing new frontiers in this concept, LIBRA AI Technologies has launched a collaborative internship with our esteemed partner ADVEOS, a leading company in microsystems technology based in Greece. This collaboration aimed to test the feasibility and assess the accuracy of the cutting-edge concept of Deep Learning-based WI-FI Sensing for real-world applications.
To foster a more profound comprehension of our endeavours' efficacy, let's start by offering concise descriptions of this advanced technology and some practical examples of how it can be used.
What is Wi-Fi Sensing with Deep Learning?
Wi-Fi Sensing leverages the existing Wi-Fi infrastructure to detect and analyse human presence and movements, even when they are not connected to the Wi-Fi network. This is achieved by utilising the changes in Wi-Fi signal patterns caused by the presence of people in particular spatial contexts. Deep Learning plays a crucial role in accurately interpreting these signal variations and extracting meaningful insights.
How does it work?
Traditional Wi-Fi systems are primarily designed for data communication between devices. However, with Wi-Fi Sensing, these systems are repurposed to act as a virtual radar for monitoring human movements. When a person moves in a Wi-Fi-covered area, their presence affects the Wi-Fi signals, causing slight disturbances. These disturbances are captured and analysed by sophisticated deep-learning algorithms.
The Role of Deep Learning:
Deep Learning algorithms, such as Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs), are trained on vast datasets to recognise and interpret the patterns in the Wi-Fi signals that correspond to specific human activities. By analysing these patterns, the Deep Learning model can infer whether someone is entering or leaving a room, walking, sitting, or engaging in other activities.
Applications and Benefits:
Wi-Fi Sensing with Deep Learning opens up a multitude of possibilities across a wide array of sectors. The following examples showcase the potential this concept has to tackle complex challenges related to critical life aspects such as health, safety and security:
- Healthcare: Monitor patient movements in hospitals or elderly care centers to ensure their safety and well-being.
- Retail Analytics: Analyse foot traffic patterns to optimise store layouts and improve customer experience.
- Security and Surveillance: Enhance security systems by detecting and tracking intruders or unusual activities.
- Industrial Safety: Monitor employee movements in hazardous environments to prevent accidents.
- Smart Home and Building Automation: Enable smart lighting and climate control systems to adjust based on occupancy, enhancing energy efficiency.
- Assisted Living: Assist people with disabilities by automating certain tasks based on their movements.
- Traffic Management: Analyse pedestrian movements in urban areas for better traffic planning.
- Automotive Safety: Monitor vehicle occupancy and alert caregivers and authorities for unsafe conditions to prevent hot car tragedies caused by several factors, such as the "forgotten baby syndrome."
The Experiment:
Guided by our CEO and Principal Data Scientist, Yannis Kopsinis, two talented interns, Serafeim Tzelepis and George Kyriakopoulos, implemented a hands-on experiment diving deep into the task of automatic people counting with deep learning techniques. The experiment involved data collection and labelling as well as the development and testing of Deep Learning algorithms for Wi-Fi Sensing.
A dedicated Wi-Fi signal receiver was used to sense a Wi-Fi signal transmitter. Both devices, provided by ADVEOS, were placed in a LIBRA AI office room to record the Wi-Fi signals in this specific working space. The signal recording was captured with a respective software developed by ADVEOS.
To label the data concerning the number of people in the room, LIBRA AI Data Scientist Christos Petrou assembled a digital counting device near the room's exit. The device consisted of a Green and a Red button and a small LED display to keep track of the counter.
An office campaign was then organised to ensure employees' participation, as it was critical for the accuracy of the developed dataset. Through this engaging and playful campaign entitled: "Hey you! Press the button; an experiment is running!", the people entering or leaving the room were asked to push the green or red buttons, increasing or decreasing the counter's number, respectively.
Sneak peek into the Wi-Fi Sensing Experiment office campaign.
This simple counting method logged the number of people in the room, providing labelled data for the supervised learning techniques. The final dataset, including these labelled data and the corresponding Wi-Fi signals data, was then used to train the developed Deep Learning models for performing two main tasks: detecting peoples' presence and counting the crowd.
Exploring the initial results
We have applied our model during a day shift after training, and the results are shown in the video below. The ground truth on how many people are in the office room across time is depicted with the dotted red line. The blue line illustrates our model's predictions. These preliminary findings demonstrate the model's ability to discern variations in room occupancy with an approximate 70% accuracy when a ±1 error is allowed.
Model Predictions vs. Ground Truth: Initial Insights
We consider these results very promising since the setup, was possibly the hardest for the people counting problem. People in an office tend not to move much and there is large variability in the places where people might be sitting. Beyond people counting, the model fully resolved the problem of detecting whether there is anyone in the room or not. While these initial outcomes are encouraging, further research is essential to refine, validate, and adapt the technology for broader real-world applications, considering the complexities and variations encountered in diverse environments and scenarios.
Final Thoughts
With continuous advancements in Artificial Intelligence technologies, Wi-Fi Sensing with Deep Learning promises to transform how we interact with our environment on several occasions. Through its ability to interpret human movements a presence in a given space, this emerging technology offers new ways to understand and engage with our surroundings, potentially revolutionising various critical aspects of our daily lives.
The practical and collaborative approach employed in this hands-on experimentation not only tested hypotheses but also stretched the boundaries of this technology, providing an invaluable learning experience for both participating companies.
LIBRA AI eagerly embraces the boundless potential of Wi-Fi Sensing with Deep Learning in our ongoing pursuit to translate innovation into tangible applications. The possibilities it presents across various sectors fuel our excitement, propelling us to explore its horizons further through the company's ongoing Research and Development activities.