Lior Weiss

People Counting: What is this good for?

‘People counting’ applications can be implemented using wireless sensing. This type of implementation is extremely useful for situations that require intelligent customer and operational insight such as retail, schools, banks, recreation facilities, museums, airports, smart building energy management, crowd control, traffic management, security, hospitality, and many others. 

Traditional solutions for people counting, such as cameras, can be expensive and have many limitations such as Line-of-Sight (LOS) requirement, good lighting conditions requirements, and privacy issues. People counting based on wireless signals (infrared, ultrasound, Wi-Fi, etc.) overcome the drawbacks of the traditional solutions since they can perform equally well in a Non-Line-of-Sight (NLOS) and low-light environment while significantly reducing privacy concerns. In particular, Wi-Fi based solutions are economical and practical since they do not require installation of any additional custom equipment. 

Orange Labs study / IEEE paper 

Orange Labs recently carried out a study of Wi-Fi based people counting and movement tracking. The results of the study were published in an IEEE paper. The study was based on the use of Celeno’s Wi-Fi Access Point (AP) with built-in Doppler Imaging for wireless sensing. 

The people counting study was conducted in a typical living room of an apartment as shown below. This room includes common furniture items such as a dining table, a small table, a bed, a closet, a desk, and chairs. A dozen different scenarios were considered including from no one being in the room to up to four people. Some of the people could be sitting while others could be moving about in the room. 

diagram-1-16 (1)

A large number of Doppler Imaging measurement samples were taken from the Celeno Wi-Fi Access Point located in the room.  

Orange Labs researchers built a 3D Convolutional Neural Network (3D-CNN) model based on Deep Learning techniques. The 3D-CNN model was trained using the Doppler Imaging data from the Celeno Wi-Fi Access Point. This model was then used for drawing inferences for various scenarios mentioned above. And the results were quite impressive and outperformed other baseline approaches. Over 89% accuracy was observed across all different scenarios including people sitting and/or walking and the number of people varying from none to up to four. 

Celeno’s Wi-Fi Doppler Imaging

Celeno’s Wi-Fi Doppler Imaging technology is the first of its kind. It harnesses Wi-Fi hardware to generate Doppler Radar Images to track objects and depict their behavior. It detects and tracks the movement and location of people, pets, and objects.

Micro-motion dynamics are typical for living objects that have limbs and joints moving in different velocities and directions. For example: when a person walks, different parts of their body move differently, and each body part generates a unique Doppler shift. Objects like cars, robots, and machines have micro-motion dynamics characterized by vibrations when moving or additional moving parts.

Doppler signature produced by the moving objects is analyzed by Celeno’s built-in machine learning classifiers, which locate, monitor, and assess the behavior of such objects and the situation itself. The technology enhances the contextual interpretation of the event, enabling an endless variety of applications. 

Celeno’s Wi-Fi Doppler Imaging enables people counting applications and beyond. For example, it has the capability to differentiate between adults, children, pets, and objects as well as distinguish between different movements such as sitting down, lying down, standing up, or falling. It can even detect breathing which is important in remote elderly care applications.  

The figure below shows an example of a Doppler Imaging data obtained from Celeno’s Wi-Fi Doppler Imaging capable devices. 

pillar-diagram-5-2The Y-axis in the figure corresponds to the Doppler frequency which relates to the amount of people movement. The figure shows the variation in the Doppler frequency as a function of time. Sudden change in the Doppler frequency could be an indication of abnormal events such as falling, as shown in the figure. While the scenario illustrated above can be interpreted by visually looking at the Doppler images, more complex scenarios require sophisticated processing. Further, the scenarios could change rapidly and the system needs to process the Doppler images fast and make quick and accurate decisions. Applying Machine Learning based solutions is the right approach for this. 


Wireless sensing and people counting is practically achievable without using any custom installation or expensive hardware. Access Points based on Celeno’s Wi-Fi solutions with Doppler Imaging technology can readily enable many different people counting applications. This approach works across different environmental conditions well in LOS and NLOS conditions and in light or dark while maintaining privacy. Celeno also offers Deep Learning software libraries to customize the solutions for particular applications. 

Learn more about Celeno’s Wi-Fi Doppler Imaging here.

Back Next


Share this post