The Portal Beam Occupancy AI Model, a Convolutional Neural Network (CNN) that resides locally on its CPU, has been trained with 350 datasets comprising over 2 million annotations of humans in office setups, including meeting rooms, seating areas, and entrance areas. This rigorous training has resulted in a highly robust prediction model, boasting an accuracy of 95-97% or higher for detecting humans in indoor environments.
The combination of the Portal Beam thermal imaging technology and its AI Model provides an additional layer of reliability and privacy compliance. Compared to non-thermal optical imaging technologies, the thermal technology + AI approach offers increased robustness in occupancy detection. Furthermore, it ensures 100% compliance with personally identifiable information (PII) and privacy regulations. From the thermal heat stamp generated by the Portal Beam thermal imaging sensor it’s impossible to identify any specific individuals.
The example below illustrates (1) the Portal Beam thermal sensor image, (2) the actual representation of people located in the space, and (3) real-time data available from Kio Cloud Smart Location.
The example below illustrates:
(1) The Portal Beam thermal sensor image
(2) The actual representation of people located in the space
(3) Real-time data available from Kio Cloud Smart Location
In addition to its accuracy and privacy features, the Portal Beam excels in energy efficiency. The Beam is equipped with an 8-core CPU with a neural network architecture that can be up to 20 times more energy efficient. This ultra-low power consumption enables efficient edge computing, allowing a Portal Beam to operate on a battery for up to 4 years with the factory default settings.
There are certain sensitivity limitations of the Portal Beam Occupancy AI Model that do exist.
In some edge cases, such as the presence of non-human heat sources or complex sunlight conditions, inaccuracies or false positives may occur. It's important to note that hot objects like fridges, laptops, monitors, TVs, and printers should not be detected by the model. Additionally, there may be instances where humans wearing umbrellas, large hats, or bulky winter jackets that cover their entire body and head may not be consistently detected.
Example of a hot object not detected by the model.
Occasionally, false positives triggered by sunlight, known as "sun ghosts," can occur. This happens when sunlight heats up surfaces like chairs, creating heat patterns that may be mistakenly interpreted as human presence.
The Kontakt.io Kio Cloud architecture is built with a data pipeline that ensures a systematic and organized flow of Portal Beam occupancy data, enabling efficient processing, analysis, and utilization of the data within the Kio Cloud Occupancy Engine.
The Portal Beam CNN (Occupancy AI Model) analyzes each thermal image in real-time to predict the count of people detected.
Once the image data is processed and analyzed, the Portal Beam discards the image locally, ensuring that no thermal images are stored on the device itself.
The CNN count of people detected, which is outputted in a numerical data format, is translated back into the Kontakt.io Telemetry eBLE packet.
This packet is then transmitted and received by a BLE to Wi-Fi infrastructure gateway device, which includes Cisco Access Points enabled by Cisco Spaces.
All data in transit and at rest is encrypted.
The Kio Cloud Occupancy Engine employs a range of analytical techniques and algorithms to aggregate the final occupancy prediction.
This aggregation is based on both the current packet data and historical data, enabling a comprehensive understanding of occupancy trends.
By default, the Occupancy Engine uses a 5-minute aggregation period, ensuring accurate occupancy data over time.
The Kio Cloud Platform stores and retains occupancy data for one year that can be retrieved through our Location and Occupancy REST API.
Learn more about the Occupancy API endpoints and data.
Alternatively, Cisco Spaces customers can benefit from a native integration via the Cisco Spaces Plugin. This allows Cisco partners and customers to consume processed Portal Beam data directly from the Cisco Spaces Meta API without the need to integrate with the Kontakt.io Location and Occupancy API separately.
Overall, the Portal Beam Occupancy Architecture, a combination of its Occupancy AI Model, thermal technology, ultra-low power consumption, and a data pipeline that delivers efficient processing and analysis of occupancy data, makes the Kontakt.io Portal Beam a robust and privacy-compliant solution for accurate occupancy detection in various indoor environments.