Advancing crowd detection analytics with deep learning and Wi-Fi

Crowd detection analytics have become increasingly accurate thanks to deep learning. Companies are even using anonymous data gathered from Wi-Fi-enabled devices to count crowds. These methods of crowd counting are providing end users with alternatives to the traditional detection, regression and/or density models.

Improving accuracy with deep learning

Machine learning has been the backbone of crowd detection analytics. Yet, as artificial intelligence has advanced, deep learning has improved the accuracy and reliability of counting crowds by taking things a step further.

With high-performance hardware readily available today, Tom Hofer, Product Manager at Senstar, pointed out that the current limitations of crowd detection analytics are typically the datasets available to train the new systems.

Incorporating deep learning mechanisms (i.e., convolutional neural networks) and increasing the size and quality of datasets has made today’s crowd detection analytics more accurate. Deep learning enables users to use large datasets and fine-tune the analytics. Furthermore, there are different datasets available for different crowd sizes.

Hofer explained, “Machine learning that adapts to the training datasets and may be limited in terms of robustness, making it harder to deploy and requiring additional inputs from the user. Pre-trained deep learning algorithms should allow users to pick a desired environment and therefore benefit from pre-trained datasets for the specific use cases.”

The challenge becomes managing environmental and crowd expectations. However, using deep-learning frameworks that decrease human inputs provides greater capabilities, which leads to higher accuracies, especially when applying analytics to larger, denser crowds.

Utilizing data from Wi-Fi networks

Crowd analytics companies are also using anonymous data from Wi-Fi enabled devices (e.g., smartphones, etc.) for crowd counting. Some researchers contend that crowd counting via Wi-Fi has advantages over video-based crowd analysis. For example, using Wi-Fi data eliminates lighting issues — in darker, low-light environments video-based crowd analytics may encounter difficulties leading to less reliable data.

Crowd counting using data detected by existing Wi-Fi networks could help organizations and venues manage social distancing, protect customers and staff, as well as develop a longer term understanding of their places.

“Deploying our solution is straightforward using major Wi-Fi access points. I’d say that the biggest advantage is that it allows organizations to be proactive in managing social distancing by knowing about and being able to reduce risks before they are a problem. Ultimately, this is about using the Wi-Fi presence data to give confidence to citizens that venue owners are taking steps to minimize risk,” said Chris Bruce, MD at GlobalReach Technology.

It is important to note that this method uses Wi-Fi network data from devices and wearables connected to venue Wi-Fi. GlobalReach Technology estimates approximately 80-percent of smartphones have Wi-Fi switched on. To account for the variability, Bruce explained that the venue is in control of the thresholds and can set alert triggers at lower thresholds.

Choosing the right solution

Ultimately it comes down to need. Will the area be densely populated? Is there available Wi-Fi access points? While there is no perfect solution for every environment or one method that will guarantee 100-percent accuracy, choosing the right solution can help organizations properly meet expectations and requirements.